Prompt_Creator_SFW_NSFW_ModelTargets

{"id":"3b758009-8c1e-443d-94e2-0a7b91055546","revision":0,"last_node_id":33,"last_link_id":24,"nodes":[{"id":1,"type":"MarkdownNote","pos":[-2600,200],"size":[520,820],"flags":{},"order":0,"mode":0,"inputs":[],"outputs":[],"title":"HOW TO USE","properties":{"Node name for S&R":"MarkdownNote","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["# Prompt Creator — SFW / NSFW · model targets\n\nLocal rewriter. No cloud API. Each TARGET node already contains that model's official prompt structure. The Qwen text model only follows the enabled TARGET text.\n\n## Use\n1. Enable **exactly one** CONTENT group (SFW or NSFW)\n2. Enable **exactly one** TARGET group\n3. Edit **USER IDEA** (add duration, first/last frame, or refs if the model needs them)\n4. Optional: enable **OPTIONAL · Ref Image** and load a still\n5. Queue Prompt — copy the result from **ENHANCED PROMPT**\n\nOutput must be the **finished generation prompt only** — no “The user wants”, no “Plan:”, no markdown fences.\n\n**H3:** paste into MiniMax H3 as-is (`integrated_multimodal_description` / `overall_soundscape` / `non_diegetic_music`, or the six Ref2VA sections).\n**LTX:** one flowing paragraph (≤200 words), action first.\n**Stills:** one cohesive visual paragraph in that model's official rewriter style.\n\nDefaults: Custom Only · max_tokens **2048** · temp **0.35** · Qwen3.5-4B-Unredacted-MAX\n\n## Ref image (16GB VRAM)\nDo **not** switch the TEXT rewriter to a VL checkpoint. Enable OPTIONAL · Ref Image. A 4B Unredacted VL reads a 768px still and writes an identity paragraph; the text 4B rewriter then builds the official-format prompt. 8B/32B VL will OOM.\n\nSee **INSTALL + DOWNLOADS** for nodes and model links.\n"],"color":"#222","bgcolor":"#000"},{"id":2,"type":"MarkdownNote","pos":[-2600,980],"size":[520,560],"flags":{},"order":1,"mode":0,"inputs":[],"outputs":[],"title":"OFFICIAL SOURCES","properties":{"Node name for S&R":"MarkdownNote","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["# Official sources baked into TARGET nodes\n\nCopied into the graph. Nothing is fetched at run time.\n\n| Target | Source |\n|---|---|\n| MiniMax H3 | [MiniMax-AI/MiniMax-H3](https://github.com/MiniMax-AI/MiniMax-H3) `skills/h3-prompt-writing` (`base-en.txt`, `ref-en.txt`) |\n| LTX-2.5 | [Lightricks/LTX-2](https://github.com/Lightricks/LTX-2) README “Prompting for LTX-2” + 2.5 multi-shot notes (also valid for 2.3) |\n| Z-Image Turbo | [Tongyi-MAI/Z-Image-Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo) prompting + official PE template (`pe.py`) |\n| Krea 2 | [krea-ai/krea-2](https://github.com/krea-ai/krea-2) `docs/prompting.md` + `docs/expansion.txt` |\n| Qwen-Image | [QwenLM/Qwen-Image](https://github.com/QwenLM/Qwen-Image) `prompt_utils_2512.py` `polish_prompt_en` |\n| Wan 2.2 | [Wan-Video/Wan2.2](https://github.com/Wan-Video/Wan2.2) user-guide formula: Subject + Scene + Motion (+ Aesthetic Control + Stylization) |\n| Flux / Flux.2 | [black-forest-labs](https://github.com/black-forest-labs) `flux-best-practices`: Subject + Action + Style + Context + Lighting + Camera |\n| SDXL / Pony | Stability SDXL natural language + Pony Diffusion score-tag convention |\n| Universal | Model-agnostic fallback |\n"],"color":"#222","bgcolor":"#000"},{"id":3,"type":"Fast Groups Bypasser (rgthree)","pos":[-1980,200],"size":[400,220],"flags":{},"order":2,"mode":0,"inputs":[],"outputs":[{"name":"OPT_CONNECTION","type":"*","links":null}],"title":"CONTENT TOGGLE — SFW / NSFW","properties":{"matchColors":"","matchTitle":"CONTENT ·","showNav":true,"showAllGraphs":true,"sort":"position","customSortAlphabet":"","toggleRestriction":"always one","nodeMode":4,"ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"color":"#322","bgcolor":"#533"},{"id":4,"type":"Fast Groups Bypasser (rgthree)","pos":[-1980,480],"size":[400,280],"flags":{},"order":3,"mode":0,"inputs":[],"outputs":[{"name":"OPT_CONNECTION","type":"*","links":null}],"title":"MODEL TOGGLE — pick one target","properties":{"matchColors":"","matchTitle":"TARGET ·","showNav":true,"showAllGraphs":true,"sort":"position","customSortAlphabet":"","toggleRestriction":"always one","nodeMode":4,"ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"color":"#232","bgcolor":"#353"},{"id":5,"type":"PrimitiveStringMultiline","pos":[-1980,820],"size":[480,360],"flags":{},"order":4,"mode":0,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[1]}],"title":"SFW","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["CONTENT RATING: SFW (Safe for Work)\n\nApply these constraints INSIDE the selected model's official output schema. Do not abandon field names, tags, or paragraph rules of the TARGET guide.\n\nHard rules:\n- No sexual content, nudity, fetish, or pornographic intent.\n- No underage characters (all people 18+ if any). Never youthful/teen wording.\n- General-audience language (G/PG-13).\n- No graphic gore, extreme violence, hate, or illegal acts.\n- Romance only if non-explicit and tasteful.\n\nRewrite USER IDEA into a generation-ready prompt that obeys the TARGET MODEL official format AND these SFW rules.\n"],"color":"#322","bgcolor":"#533"},{"id":6,"type":"PrimitiveStringMultiline","pos":[-1460,820],"size":[480,360],"flags":{},"order":5,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[2]}],"title":"NSFW","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["CONTENT RATING: NSFW (Adult / explicit allowed)\n\nApply these constraints INSIDE the selected model's official output schema. Do not abandon field names, tags, or paragraph rules of the TARGET guide.\n\nHard rules:\n- Adult content only for clearly 18+ subjects. Never minors or youthful wording.\n- If the user wants explicit content, be concrete and anatomically coherent (body, acts, clothing state, consenting adults).\n- Prefer cinematic / photographic / official-guide camera language over crude slang unless the user used slang.\n- Keep identity, lighting, camera, and continuity coherent.\n- Never invent illegal content.\n\nRewrite USER IDEA into a generation-ready prompt that obeys the TARGET MODEL official format AND these NSFW rules.\n"],"color":"#322","bgcolor":"#533"},{"id":7,"type":"PrimitiveStringMultiline","pos":[-1980,1300],"size":[520,360],"flags":{},"order":6,"mode":0,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[3]}],"title":"MiniMax H3 (official T2VA/I2VA/Ref2VA)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: MiniMax H3\nOfficial guide baked in from MiniMax-AI/MiniMax-H3 skills/h3-prompt-writing (base-en.txt + ref-en.txt). Do not fetch URLs. Follow this structure exactly.\n\nYou write ONE finished H3 prompt. Output ONLY that prompt. English body. Preserve dialogue, lyrics, and on-screen text in their original language, verbatim.\n\nMODE (choose from USER IDEA; default T2VA):\n- T2VA: text only.\n- I2VA: first frame / I2V / opening still / <Picture 1> as first frame.\n- FL2VA: first AND last frame.\n- L2VA: last frame only.\n- Ref2VA: R2V / identity or style refs / reference video or audio / <Picture>, <Video>, <Audio>.\n\nDuration: use the user's duration; if missing assume 5 seconds. Stay inside 4–15 seconds. Do not add a timestamp to [Shot 1]. Later shots: [Shot N] At MM:SS.mmm,  with strictly increasing times inside the duration.\n\n========================\nBASE MODES (T2VA/I2VA/FL2VA/L2VA)\n========================\nI2VA must start with this exact instruction then a blank line:\nFor the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.\n\nFL2VA must start with:\nHow the reference pictures align with the target video — Picture 1 (from Shot 1) aligns with the 0.00-second mark of the target video; Picture 2 (from Shot N) aligns with the S.SS-second mark of the target video.\n\nL2VA must start with:\nHow the reference pictures align with the target video — <Picture 1> (from [Shot N]) aligns with the S.SS-second mark of the target video.\n\nThen EXACTLY these three fields, in this order, with these field names:\n\nintegrated_multimodal_description: [Shot 1] Live-action, cinematic, a {shot size} frames {subjects, composition, scene}. {camera motion as a natural sentence}. {speaker (S1) action} says: <d>[English] {verbatim line}</d> [Shot 2] At 00:05.000, the camera cuts to ...\n\noverall_soundscape: 1–4 English sentences, one paragraph: ambient + physical action + non-verbal human sound (rain, footsteps, fabric, breath). Do NOT repeat dialogue or singing. Use N/A only if the user asked for complete silence.\n\nnon_diegetic_music: 1–3 English sentences: instrumentation, tempo, rhythm, dynamics. No abstract mood words. Diegetic radio/TV/phone music belongs in integrated_multimodal_description. N/A if none.\n\nSHOT RULES (official):\n- [Shot 1] opens with style + composition. Styles: Cinematic, live-action, 2D-animated, 3D CG, claymation, watercolor, vintage film.\n- Prefer camera motion over extra cuts. Cut phrases: the camera cuts to / the shot cuts to / the shot transitions to.\n- Camera motion = type + optional amplitude + optional speed, written as a sentence:\n  Zoom In/Out, Push In/Pull Out, Pan Left/Right, Truck Left/Right, Tilt Up/Down, Pedestal Up/Down, Arc Shot, Tracking Shot, Static Shot, Shake Slightly/Strongly, POV, Roll Clockwise/Counterclockwise.\n  Amplitude: with small amplitude / with large amplitude. Speed: at slow speed / at fast speed.\n  Example: The camera pushes in with small amplitude at slow speed toward the folded letter in her hands.\n- Speakers: stable (S1), (S2). Identify type/age/gender/timbre on first vocal event. Dialogue and lyrics ONLY inside <d>[Language] ...</d>. Do not translate or rewrite user words. Voiceover: says in an off-screen voiceover: <d>...</d> while lips remain completely closed.\n- Visible sign/banner/subtitle text in English double quotes, verbatim.\n- Concrete visible/audible detail. Do not output [0s-2s] blocks. Do not output \"The user wants\".\n\nI2VA: Picture 1 is the actual first frame at 0.00s in [Shot 1]. Anchor style/subjects/composition from the still, then develop action. Keep identity, clothing, colors, props, spatial relations.\nFL2VA: describe the path from Picture 1 to Picture 2; prefer a single shot unless the user asked for cuts.\nL2VA: infer a plausible opening and land on Picture 1 at the end.\n\n========================\nREF2VA (full reference)\n========================\nIf USER IDEA is reference-to-video, output these six sections IN THIS ORDER with these names:\n\nsubject_definitions:\nsummary:\nretention_analysis:\ndetailed_description:\noverall_soundscape:\nnon_diegetic_music:\n\nLabels (keep consistent in every section):\n<Subject N> reusable visible content (person, clothes, environment, style)\n<Picture N> image used as a concrete frame/keyframe/composition anchor\n<Video N> source video for edit/continuation/structure\n<Audio N> audio copy or reference\n\nsummary starts with a square-bracket task type, e.g. [reference generation] or [video editing + audio reuse]. Types: keyframe completion, reference generation, video editing, video continuation, audio reuse, audio reference. Combine with \" + \".\nretention_analysis: one line per label.\nVisual markers: fully_preserved | partially_preserved | attribute_transfer | weak_reference\nAudio markers: fully_copy | partially_copy | reference | weak_reference\nExample: <Subject 1> (appears in [Shot 1]): fully_preserved - ...\ndetailed_description: 1–2 English style sentences, then [Shot 1] with no timestamp. Insert labels where they apply. For generation, aim 350–500 words unless dialogue-dense. Speaker form: <Subject N> (S1) says, <d>[English] ...</d>\n\nDo not invent a different story. Preserve must-keep keywords. All people are consenting adults 18+. Finish the full prompt; never stop mid-sentence.\n"],"color":"#232","bgcolor":"#353"},{"id":8,"type":"PrimitiveStringMultiline","pos":[-1420,1300],"size":[520,360],"flags":{},"order":7,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[4]}],"title":"LTX-2.5 (official LTX-2 prompting)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: LTX-2.5 (official LTX-2 prompting; also valid for LTX-2.3)\nOfficial guide baked in from github.com/Lightricks/LTX-2 README “Prompting for LTX-2” and LTX-2.5 multi-shot notes. Do not fetch URLs.\n\nWrite ONE finished LTX prompt. Output ONLY that prompt. No preface, no markdown, no shot-list bullets unless you also describe the cut in prose.\n\nOFFICIAL SINGLE-SHOT STRUCTURE (default):\nOne flowing paragraph, chronological, literal, precise. Start DIRECTLY with the action. Think like a cinematographer describing a shot. KEEP WITHIN 200 WORDS.\n\nBuild in this order:\n1) Main action in a single sentence\n2) Specific details about movements and gestures\n3) Character/object appearances precisely\n4) Background and environment details\n5) Camera angles and movements\n6) Lighting and colors\n7) Any changes or sudden events\n\nRules:\n- Natural cinematic language (tracking shot, shallow depth of field, golden hour, low angle, macro lens).\n- No SD weight syntax like (word:1.3). No tag soup. No [0s-2s] blocks.\n- Include native audio in the same paragraph when the user wants sound: who speaks, exact quoted dialogue, SFX, ambience.\n- Prefer one continuous take (camera moves, not edits) unless the user asked for cuts.\n\nLTX-2.5 MULTI-SHOT (only if the user asked for cuts / multiple shots, 2–4):\nStill chronological prose. Name the edit (hard cut, match cut, dissolve). Re-identify subjects after each cut. At every cut, say whether music / dialogue / ambience continues or changes. Do not use screenplay sluglines unless the cut is also described in prose.\n\nDomain vocabulary is allowed when it fits (VFX: rack focus, rotoscope; animation: ease-in/ease-out; restoration: gate weave) but do not stack equal-weight styles — pick a primary.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":9,"type":"PrimitiveStringMultiline","pos":[-860,1300],"size":[520,360],"flags":{},"order":8,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[5]}],"title":"Z-Image Turbo (official PE)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: Z-Image Turbo (Tongyi-MAI)\nOfficial guidance baked in from Tongyi-MAI/Z-Image-Turbo prompting notes and the official PE template (pe.py). Do not fetch URLs.\n\nWrite ONE finished still-image prompt. Output ONLY that prompt. No preface, no markdown, no Negative: section.\n\nOFFICIAL PE RULES:\n- You are rewriting the user prompt into a faithful, detail-rich, aesthetic visual description a text-to-image model can use directly.\n- LOCK unchanged core elements first: subject, count, action, state, IP names, specified colors, specified text.\n- Then add professional aesthetic/realism detail: composition, lighting, materials, color, spatial layers.\n- Objective and concrete. Forbidden: metaphor, emotional rhetoric, meta tags such as \"8K\", \"masterpiece\", \"best quality\", \"ultra HD\".\n- If any text must appear in the image, transcribe it EXACTLY inside English double quotes (\"like this\"), and describe placement, size, and material.\n- Long, detailed natural-language sentences. Not a booru tag list. Not a video storyboard.\n- Z-Image-Turbo uses NO negative prompts and NO classifier-free guidance. Put exclusions as positive facts (\"plain background, no watermark\") only if the user needs them, never as a Negative block.\n\nCover: shot/subject, appearance, clothing, environment, lighting, camera/lens, medium/style, grounded spatial relations.\n\nPreserve must-keep keywords. All people 18+. One continuous visual description.\n"],"color":"#232","bgcolor":"#353"},{"id":10,"type":"PrimitiveStringMultiline","pos":[-1980,1720],"size":[520,360],"flags":{},"order":9,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[6]}],"title":"Krea 2 (official expansion.txt)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: Krea 2 (official krea-ai/krea-2)\nOfficial guidance baked in from github.com/krea-ai/krea-2 docs/prompting.md and docs/expansion.txt. Do not fetch URLs.\n\nWrite ONE finished still-image prompt. Output ONLY a single expanded paragraph. No preface, no bullets, no JSON, no markdown.\n\nOFFICIAL EXPANSION RULES:\n1. Faithfulness first: preserve original subjects, actions, colors, spatial relationships. Do not add new objects, props, characters, or animals unless clearly implied.\n2. Practical T2I structure: group subjects with their own attributes and actions. Grounded poses, interactions, layout.\n3. Choose style/medium/framing/lighting internally; do not emit planning tags.\n4. Visible text: wrap requested words in quotes and specify placement.\n5. Avoid over-specification: do not invent highly specific clothing, colors, or materials the input does not support.\n6. One cohesive paragraph. If the user prompt is already detailed, lightly polish rather than heavily expand.\n7. Preserve the user's medium (\"photo of\", \"illustration of\", \"3D render of\").\n8. Natural language. Long detailed prompts yield best controlled results; short prompts are OK only if the user idea is already a complete image.\n9. No SD1.5 quality spam (\"masterpiece, best quality, 8k\") and no (word:1.4) weights.\n\nUseful order from official examples: subject, setting, composition, lighting, mood, medium/style, texture/camera.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":11,"type":"PrimitiveStringMultiline","pos":[-1420,1720],"size":[520,360],"flags":{},"order":10,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[7]}],"title":"Qwen-Image (official polish_prompt_en)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: Qwen-Image / Qwen-Image-Edit\nOfficial rewriter baked in from QwenLM/Qwen-Image src/examples/tools/prompt_utils_2512.py polish_prompt_en. Do not fetch URLs.\n\nWrite ONE finished English image prompt. Output ONLY that prompt as a single continuous block. No markdown lists, no numbered headings, no preface.\n\nClassify the request as portrait, text-containing image, or general image, then rewrite.\n\nCORE REQUIREMENTS (official):\n1. Fluent natural descriptive language. No bullet lists. Line breaks OK only for infographic sections.\n2. Enrich visual details that stay consistent with the original. Do not alter original concepts. Do not add text to images that should have none.\n3. Never modify proper nouns (names, brands, places, IPs, slogans, URLs).\n4. Fully represent visible text: every displayed string in English double quotes (\" \"). Describe position, layout direction, font, color, size, presentation (printed/neon/embroidered). If no text, state: The image contains no recognizable text.\n5. Clearly specify overall artistic style (realistic photography, anime illustration, movie poster, watercolor, 3D rendering, game CG, etc.).\n\nPORTRAIT: ethnicity, gender, specific age or narrow age range (not just \"young\"), face shape, eyes/nose/mouth, skin/makeup, clothing with fabric, hairstyle, accessories, pose, gaze, hands, background, lighting. Aim ~200 words. Anatomically plausible.\n\nTEXT-CONTAINING: faithful transcription, carrier (sign/screen/poster), lighting vs readability. Do not use vague \"a list\"; give concrete quoted strings if the user implied text.\n\nGENERAL (landscape/still life): subject form/color/material, foreground/mid/background, lighting, textures, weather, scale.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":12,"type":"PrimitiveStringMultiline","pos":[-860,1720],"size":[520,360],"flags":{},"order":11,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[8]}],"title":"Wan 2.2 (official formula)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: Wan 2.2 (Wan-Video official formula)\nOfficial guidance baked in from the Wan2.2 user-guide prompt formulas. Do not fetch URLs.\n\nWrite ONE finished Wan 2.2 video prompt. Output ONLY that prompt. No preface, no markdown fences, no SD weights.\n\nOFFICIAL FORMULAS:\nBasic: Subject + Scene + Motion\nAdvanced (default): Subject (Description) + Scene (Description) + Motion (Description) + Aesthetic Control + Stylization\n\nUse the advanced formula unless the user idea is a one-line sketch.\n\nSubject: who/what, appearance, wardrobe, expression.\nScene: environment, foreground and background, time of day, weather.\nMotion: what moves, how, speed; include camera motion as part of Motion.\nAesthetic Control (pick a compact set, 2–6 tags, consistent): light source (daylight, moonlight, practical light, neon), light quality (soft light, hard light, rim light), time of day, shot size (wide / medium / close-up), composition (centered, left-weighted, rule of thirds), lens (wide angle, 35mm, telephoto), angle (eye level, low angle, high angle, over-the-shoulder, POV).\nStylization: cinematic live-action, anime, 3D, documentary, etc. — one primary.\n\nShot order: Opening shot → Camera motion → Pay-off, as one coherent clip unless the user asked for multiple shots.\nCamera verbs: pan, tilt, dolly, orbit, crane, tracking, static.\nI2V: say the subject matches the reference still (identity, outfit, framing) then describe only the motion that happens next.\nWan open-weights is visual-motion first; a short ambience line is optional, not a MiniMax-style audio script.\nLength: dense but not a trailer. About 80–120 words is the official-style working range.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":13,"type":"PrimitiveStringMultiline","pos":[-1980,2140],"size":[520,360],"flags":{},"order":12,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[9]}],"title":"Flux / Flux.2 (official BFL formula)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: Flux / Flux.2 (Black Forest Labs)\nOfficial guidance baked in from black-forest-labs flux-best-practices. Do not fetch URLs.\n\nWrite ONE finished Flux-family image prompt. Output ONLY that prompt. No preface, no markdown, no Negative: block.\n\nOFFICIAL FORMULA:\n[Subject] + [Action/Pose] + [Style/Medium] + [Context/Setting] + [Lighting] + [Camera/Technical]\n\nCRITICAL RULES:\n1. NO negative prompts. Flux does not use them. Describe what you want, not what to avoid (unless the user insisted, then phrase as a positive scene fact).\n2. Be specific. Vague prompts produce mediocre results.\n3. Natural language / prose. Not booru tags. Not (word:1.2) weights. Not \"masterpiece, 8k\" spam.\n4. Specify lighting — it has the biggest quality impact (direction, quality, practical sources, color temperature).\n5. Quote on-image text: \"quoted text\" for typography.\n6. Hex colors when a brand/color must be exact: #RRGGBB plus a plain-language color name.\n7. Photoreal: skin/material texture, lens, practical lights. Design/layout: exact quoted copy and placement.\n8. One dense paragraph unless the user asked for multi-panel.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":14,"type":"PrimitiveStringMultiline","pos":[-1420,2140],"size":[520,360],"flags":{},"order":13,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[10]}],"title":"SDXL / Pony (Stability + Pony scores)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: SDXL / Pony\nOfficial-style guidance: Stability SDXL (natural language into CLIP-L + OpenCLIP-G) and Pony Diffusion's published score-tag convention. Do not fetch URLs.\n\nWrite ONE finished image prompt. Output ONLY that prompt.\n\nDetect style from USER IDEA.\n\nIF photoreal / general SDXL (default):\n- Natural language, subject first, then appearance, clothing, pose, environment, lighting, camera, style.\n- Light quality tokens at the end only if useful: highly detailed, sharp focus. Do NOT dump 20 quality tags. Avoid (word:1.4) unless the user writes weights.\n- Optional short last line: Negative: blurry, extra fingers, watermark  — only if helpful.\n\nIF anime / Pony / \"pony\" / score tags requested:\nUse Pony's conventional prefix then the scene:\nscore_9, score_8_up, score_7_up, source_pony, rating_safe, 1girl, {danbooru-like tags for body/clothes/pose}, {short natural-language scene}\n- SFW content → rating_safe. NSFW content (only if CONTENT is NSFW) → rating_explicit. Never rating that implies underage.\n- Keep the tag block readable; do not explode into 80 tags.\n- Then a short natural-language sentence for setting and lighting.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":15,"type":"PrimitiveStringMultiline","pos":[-860,2140],"size":[520,360],"flags":{},"order":14,"mode":4,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[11]}],"title":"Universal Detail","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["TARGET MODEL: Universal high-detail (fallback; no single official repo)\nUse only when the user did not pick a specialist model.\n\nWrite ONE finished generation prompt (still or video matching USER IDEA). Output ONLY that prompt.\n\nCover: subject, action, environment, lighting, camera/composition, materials/textures, color, mood.\nConcrete and filmable. No contradictory styles. No sampler/seed/CFG. No \"prompt:\" labels.\nOne or two tight paragraphs.\n\nIf USER IDEA is clearly video, describe chronological action and camera in prose (not H3 fields, not LTX 200-word limit unless it still reads well).\nIf still, one visual paragraph.\n\nPreserve must-keep keywords. All people 18+.\n"],"color":"#232","bgcolor":"#353"},{"id":16,"type":"Any Switch (rgthree)","pos":[-1980,2620],"size":[320,120],"flags":{},"order":23,"mode":0,"inputs":[{"dir":3,"name":"any_01","type":"STRING","link":1},{"dir":3,"name":"any_02","type":"STRING","link":2},{"dir":3,"name":"any_03","type":"STRING","link":null},{"dir":3,"name":"any_04","type":"STRING","link":null},{"dir":3,"name":"any_05","type":"STRING","link":null}],"outputs":[{"dir":4,"label":"STRING","name":"*","type":"STRING","links":[12]}],"title":"Active Content Rules","properties":{"Node name for S&R":"Any Switch (rgthree)","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":[],"color":"#233","bgcolor":"#355"},{"id":17,"type":"Any Switch (rgthree)","pos":[-1600,2620],"size":[320,206],"flags":{},"order":24,"mode":0,"inputs":[{"dir":3,"name":"any_01","type":"STRING","link":3},{"dir":3,"name":"any_02","type":"STRING","link":4},{"dir":3,"name":"any_03","type":"STRING","link":5},{"dir":3,"name":"any_04","type":"STRING","link":6},{"dir":3,"name":"any_05","type":"STRING","link":7},{"dir":3,"name":"any_06","type":"STRING","link":8},{"dir":3,"name":"any_07","type":"STRING","link":9},{"dir":3,"name":"any_08","type":"STRING","link":10},{"dir":3,"name":"any_09","type":"STRING","link":11},{"dir":3,"name":"any_10","type":"STRING","link":null}],"outputs":[{"dir":4,"label":"STRING","name":"*","type":"STRING","links":[13]}],"title":"Active Model Profile","properties":{"Node name for S&R":"Any Switch (rgthree)","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":[],"color":"#233","bgcolor":"#355"},{"id":18,"type":"JoinStrings","pos":[-1200,2620],"size":[360,100],"flags":{},"order":26,"mode":0,"inputs":[{"localized_name":"string1","name":"string1","shape":7,"type":"STRING","link":12},{"localized_name":"string2","name":"string2","shape":7,"type":"STRING","link":13},{"localized_name":"delimiter","name":"delimiter","type":"STRING","widget":{"name":"delimiter"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[15]}],"title":"Build System Prompt","properties":{"Node name for S&R":"JoinStrings","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["\n\n"],"color":"#233","bgcolor":"#355"},{"id":20,"type":"LoadImage","pos":[-800,560],"size":[360,340],"flags":{},"order":16,"mode":4,"inputs":[{"localized_name":"image","name":"image","type":"COMBO","widget":{"name":"image"},"link":null},{"localized_name":"choose file to upload","name":"upload","type":"IMAGEUPLOAD","widget":{"name":"upload"},"link":null}],"outputs":[{"localized_name":"IMAGE","name":"IMAGE","type":"IMAGE","links":[18]},{"localized_name":"MASK","name":"MASK","type":"MASK","links":[]}],"title":"OPTIONAL Ref Image","properties":{"Node name for S&R":"LoadImage","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["example.png","image"],"color":"#323","bgcolor":"#535"},{"id":22,"type":"AILab_QwenVL_PromptEnhancer","pos":[-240,200],"size":[520,520],"flags":{},"order":30,"mode":0,"inputs":[{"localized_name":"image","name":"image","shape":7,"type":"IMAGE","link":null},{"localized_name":"model_name","name":"model_name","type":"COMBO","widget":{"name":"model_name"},"link":null},{"localized_name":"quantization","name":"quantization","type":"COMBO","widget":{"name":"quantization"},"link":null},{"localized_name":"attention_mode","name":"attention_mode","type":"COMBO","widget":{"name":"attention_mode"},"link":null},{"localized_name":"use_torch_compile","name":"use_torch_compile","type":"BOOLEAN","widget":{"name":"use_torch_compile"},"link":null},{"localized_name":"device","name":"device","type":"COMBO","widget":{"name":"device"},"link":null},{"localized_name":"prompt_text","name":"prompt_text","type":"STRING","widget":{"name":"prompt_text"},"link":24},{"localized_name":"enhancement_style","name":"enhancement_style","type":"COMBO","widget":{"name":"enhancement_style"},"link":null},{"localized_name":"custom_system_prompt","name":"custom_system_prompt","type":"STRING","widget":{"name":"custom_system_prompt"},"link":15},{"localized_name":"max_tokens","name":"max_tokens","type":"INT","widget":{"name":"max_tokens"},"link":null},{"localized_name":"temperature","name":"temperature","type":"FLOAT","widget":{"name":"temperature"},"link":null},{"localized_name":"top_p","name":"top_p","type":"FLOAT","widget":{"name":"top_p"},"link":null},{"localized_name":"repetition_penalty","name":"repetition_penalty","type":"FLOAT","widget":{"name":"repetition_penalty"},"link":null},{"localized_name":"keep_model_loaded","name":"keep_model_loaded","type":"BOOLEAN","widget":{"name":"keep_model_loaded"},"link":null},{"localized_name":"seed","name":"seed","type":"INT","widget":{"name":"seed"},"link":null},{"localized_name":"keep_last_prompt","name":"keep_last_prompt","type":"BOOLEAN","widget":{"name":"keep_last_prompt"},"link":null}],"outputs":[{"localized_name":"ENHANCED_OUTPUT","name":"ENHANCED_OUTPUT","type":"STRING","links":[16]}],"title":"TEXT rewriter (no image · official guide)","properties":{"Node name for S&R":"AILab_QwenVL_PromptEnhancer","models":[{"name":"Qwen3.5-4B-Unredacted-MAX","url":"https://huggingface.co/prithivMLmods/Qwen3.5-4B-Unredacted-MAX","directory":"LLM/Qwen-VL"}],"cnr_id":"ComfyUI-QwenVL","ue_properties":{"widget_ue_connectable":{"prompt_text":true,"custom_system_prompt":true},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["Qwen3.5-4B-Unredacted-MAX","8-bit (Balanced)","sdpa",false,"auto","","✍️ Custom Only (no preset)","",2048,0.35,0.9,1.08,true,1,false,false],"color":"#233","bgcolor":"#355"},{"id":23,"type":"ShowText|pysssss","pos":[340,200],"size":[560,400],"flags":{},"order":31,"mode":0,"inputs":[{"localized_name":"text","name":"text","type":"STRING","link":16}],"outputs":[{"localized_name":"STRING","name":"STRING","shape":6,"type":"STRING","links":[]}],"title":"ENHANCED PROMPT (copy this)","properties":{"Node name for S&R":"ShowText|pysssss","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":[],"color":"#323","bgcolor":"#535"},{"id":24,"type":"PrimitiveStringMultiline","pos":[340,640],"size":[560,280],"flags":{},"order":18,"mode":0,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[]}],"title":"Clipboard helper / notes","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["Queue Prompt. Copy the Show Text box above.\n\nH3 should start with integrated_multimodal_description: or an alignment line / subject_definitions:\nLTX should be one paragraph starting with the action.\nStills should be one visual paragraph, no Negative: unless SDXL."],"color":"#222","bgcolor":"#000"},{"id":25,"type":"MarkdownNote","pos":[-240,760],"size":[520,280],"flags":{},"order":19,"mode":0,"inputs":[],"outputs":[],"title":"QUALITY KNOBS","properties":{"Node name for S&R":"MarkdownNote","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["## Enhancer knobs\n\nH3 and Qwen-Image rewrites are long. Keep max_tokens **2048**.\n\n| Goal | temp | max_tokens | quant |\n|---|---:|---:|---|\n| Strict official format | 0.30–0.40 | 1536–2048 | 8-bit |\n| Creative expand (stills) | 0.5–0.7 | 1024–1536 | 8-bit |\n| Long H3 Ref2VA | 0.35 | 2048 | 8-bit |\n\nStyle = **Custom Only**. The baked TARGET text is the system prompt and must win.\n"],"color":"#222","bgcolor":"#000"},{"id":27,"type":"ImageScaleToMaxDimension","pos":[-800,920],"size":[320,82],"flags":{},"order":25,"mode":4,"inputs":[{"localized_name":"image","name":"image","type":"IMAGE","link":18},{"localized_name":"upscale_method","name":"upscale_method","type":"COMBO","widget":{"name":"upscale_method"},"link":null},{"localized_name":"largest_size","name":"largest_size","type":"INT","widget":{"name":"largest_size"},"link":null}],"outputs":[{"localized_name":"IMAGE","name":"IMAGE","type":"IMAGE","links":[19]}],"title":"Scale ref to 768px (VRAM)","properties":{"Node name for S&R":"ImageScaleToMaxDimension","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["lanczos",768],"color":"#323","bgcolor":"#535"},{"id":29,"type":"AILab_QwenVL_Advanced","pos":[-800,1030],"size":[400,492],"flags":{},"order":27,"mode":4,"inputs":[{"localized_name":"image","name":"image","shape":7,"type":"IMAGE","link":19},{"localized_name":"image2","name":"image2","shape":7,"type":"IMAGE","link":null},{"localized_name":"model_name","name":"model_name","type":"COMBO","widget":{"name":"model_name"},"link":null},{"localized_name":"quantization","name":"quantization","type":"COMBO","widget":{"name":"quantization"},"link":null},{"localized_name":"attention_mode","name":"attention_mode","type":"COMBO","widget":{"name":"attention_mode"},"link":null},{"localized_name":"use_torch_compile","name":"use_torch_compile","type":"BOOLEAN","widget":{"name":"use_torch_compile"},"link":null},{"localized_name":"device","name":"device","type":"COMBO","widget":{"name":"device"},"link":null},{"localized_name":"preset_prompt","name":"preset_prompt","type":"COMBO","widget":{"name":"preset_prompt"},"link":null},{"localized_name":"custom_prompt","name":"custom_prompt","type":"STRING","widget":{"name":"custom_prompt"},"link":null},{"localized_name":"max_tokens","name":"max_tokens","type":"INT","widget":{"name":"max_tokens"},"link":null},{"localized_name":"temperature","name":"temperature","type":"FLOAT","widget":{"name":"temperature"},"link":null},{"localized_name":"top_p","name":"top_p","type":"FLOAT","widget":{"name":"top_p"},"link":null},{"localized_name":"num_beams","name":"num_beams","type":"INT","widget":{"name":"num_beams"},"link":null},{"localized_name":"repetition_penalty","name":"repetition_penalty","type":"FLOAT","widget":{"name":"repetition_penalty"},"link":null},{"localized_name":"frame_count","name":"frame_count","type":"INT","widget":{"name":"frame_count"},"link":null},{"localized_name":"keep_model_loaded","name":"keep_model_loaded","type":"BOOLEAN","widget":{"name":"keep_model_loaded"},"link":null},{"localized_name":"seed","name":"seed","type":"INT","widget":{"name":"seed"},"link":null},{"localized_name":"keep_last_prompt","name":"keep_last_prompt","type":"BOOLEAN","widget":{"name":"keep_last_prompt"},"link":null}],"outputs":[{"localized_name":"RESPONSE","name":"RESPONSE","type":"STRING","links":[21]}],"title":"VL identity (4B Unredacted · 768px)","properties":{"Node name for S&R":"AILab_QwenVL_Advanced","models":[{"name":"Qwen3-VL-4B-Instruct-Unredacted-MAX","url":"https://huggingface.co/prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX","directory":"LLM/Qwen-VL"}],"cnr_id":"ComfyUI-QwenVL","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["Qwen3-VL-4B-Instruct-Unredacted-MAX","None (FP16)","sdpa",false,"auto","Describe this image in detail.","Describe this still in one dense paragraph for a later prompt rewriter. Lock identity: adult 18+, face, body type, skin, hair, pose, clothing or nudity, setting, lighting. If the image is sexual, describe visible anatomy and acts concretely. If not, stay non-explicit. Do not write a MiniMax/LTX/Flux prompt. Do not say The user wants. Output only the description.",384,0.4,0.9,1,1.08,1,false,1,false,false],"color":"#323","bgcolor":"#535"},{"id":30,"type":"PrimitiveStringMultiline","pos":[-800,2620],"size":[280,88],"flags":{},"order":21,"mode":0,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[22]}],"title":"No-ref fallback (empty)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":[""],"color":"#222","bgcolor":"#000"},{"id":31,"type":"Any Switch (rgthree)","pos":[-480,2620],"size":[280,106],"flags":{},"order":28,"mode":0,"inputs":[{"dir":3,"name":"any_01","type":"STRING","link":21},{"dir":3,"name":"any_02","type":"STRING","link":22},{"name":"any_03","type":"STRING","link":null},{"name":"any_04","type":"STRING","link":null},{"name":"any_05","type":"STRING","link":null}],"outputs":[{"dir":4,"label":"STRING","name":"*","type":"STRING","links":[23]}],"title":"Identity paragraph or empty","properties":{"Node name for S&R":"Any Switch (rgthree)","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":[],"color":"#233","bgcolor":"#355"},{"id":32,"type":"JoinStrings","pos":[-800,2740],"size":[360,100],"flags":{},"order":29,"mode":0,"inputs":[{"localized_name":"string1","name":"string1","shape":7,"type":"STRING","link":14},{"localized_name":"string2","name":"string2","shape":7,"type":"STRING","link":23},{"localized_name":"delimiter","name":"delimiter","type":"STRING","widget":{"name":"delimiter"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[24]}],"title":"User idea + identity","properties":{"Node name for S&R":"JoinStrings","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["\n\nREFERENCE STILL (ignore this heading if nothing follows):\n"],"color":"#233","bgcolor":"#355"},{"id":33,"type":"MarkdownNote","pos":[-2600,1580],"size":[520,980],"flags":{},"order":22,"mode":0,"inputs":[],"outputs":[],"title":"INSTALL + DOWNLOADS","properties":{"Node name for S&R":"MarkdownNote","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["# Install + downloads\n\nAll custom nodes below are **public**. None are private or local-only.\n\n## Custom nodes (ComfyUI Manager or git)\n\n| Pack | Nodes used | Link |\n|---|---|---|\n| **ComfyUI-QwenVL** | Prompt Enhancer, QwenVL Advanced | [1038lab/ComfyUI-QwenVL](https://github.com/1038lab/ComfyUI-QwenVL) |\n| **rgthree-comfy** | Fast Groups Bypasser, Any Switch | [rgthree/rgthree-comfy](https://github.com/rgthree/rgthree-comfy) |\n| **KJNodes** | JoinStrings | [kijai/ComfyUI-KJNodes](https://github.com/kijai/ComfyUI-KJNodes) |\n| **ComfyUI-Custom-Scripts** | Show Text | [pythongosssss/ComfyUI-Custom-Scripts](https://github.com/pythongosssss/ComfyUI-Custom-Scripts) |\n\n`ImageScaleToMaxDimension`, `MarkdownNote`, `LoadImage`, `PrimitiveStringMultiline` are ComfyUI core. Update ComfyUI if any of those are missing.\n\nUnredacted-MAX is **not** in stock 1038lab `hf_models.json`. Either:\n\n- Use a public fork that already lists them: [huchukato/ComfyUI-QwenVL-Mod](https://github.com/huchukato/ComfyUI-QwenVL-Mod) or [Deaquay/ComfyUI-Qwen3.5-Uncensored](https://github.com/Deaquay/ComfyUI-Qwen3.5-Uncensored)\n- Or keep 1038lab and drop the bundled `custom_models.json` into `ComfyUI/custom_nodes/ComfyUI-QwenVL/`, then restart\n\nManager: search each pack name → Install → restart ComfyUI.\n\n```\ncd ComfyUI/custom_nodes\ngit clone https://github.com/1038lab/ComfyUI-QwenVL.git\ngit clone https://github.com/rgthree/rgthree-comfy.git\ngit clone https://github.com/kijai/ComfyUI-KJNodes.git\ngit clone https://github.com/pythongosssss/ComfyUI-Custom-Scripts.git\n```\n\n## Models (auto-download on first Queue if listed)\n\nPlace Hugging Face repos under `ComfyUI/models/LLM/Qwen-VL/<folder-name>/`.\n\n**Required — text rewriter**\n[prithivMLmods/Qwen3.5-4B-Unredacted-MAX](https://huggingface.co/prithivMLmods/Qwen3.5-4B-Unredacted-MAX)\n→ `models/LLM/Qwen-VL/Qwen3.5-4B-Unredacted-MAX/`\n\n**Optional — ref-image identity (enable OPTIONAL · Ref Image)**\n[prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX](https://huggingface.co/prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX)\n→ `models/LLM/Qwen-VL/Qwen3-VL-4B-Instruct-Unredacted-MAX/`\n\nFirst Queue Prompt will also pull them if the QwenVL pack can see those names.\n\n**Stock (censored) fallbacks if you skip Unredacted**\n[Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B) on the TEXT rewriter\n[Qwen/Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) on the VL identity node\n\nNSFW TARGET output needs the Unredacted pair. Official Qwen will refuse or soften explicit USER IDEA.\n\n## Disk / VRAM\n- Text 4B 8-bit: roughly 4 GB VRAM, ~8 GB disk\n- VL 4B FP16: roughly 6 GB extra if you enable ref image; use 4-bit if OOM\n- Do not load 8B/32B VL on 16 GB\n"],"color":"#222","bgcolor":"#000"},{"id":26,"type":"Fast Groups Bypasser (rgthree)","pos":[-1537.4524388378393,231.33133998772863],"size":[400,160],"flags":{},"order":20,"mode":0,"inputs":[],"outputs":[{"name":"OPT_CONNECTION","type":"*","links":null}],"title":"OPTIONAL REF IMAGE toggle","properties":{"matchColors":"","matchTitle":"OPTIONAL ·","showNav":true,"showAllGraphs":true,"sort":"position","customSortAlphabet":"","toggleRestriction":"default","nodeMode":4,"ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"color":"#323","bgcolor":"#535"},{"id":21,"type":"MarkdownNote","pos":[-1235.7854749077187,528.8712348114567],"size":[360,220],"flags":{},"order":17,"mode":0,"inputs":[],"outputs":[],"title":"Image tip","properties":{"Node name for S&R":"MarkdownNote","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["### Optional reference image (16GB)\n\n1. Enable **OPTIONAL · Ref Image**\n2. Load the still — it is scaled to 768px on the long side\n3. Leave the TEXT rewriter on **Qwen3.5-4B-Unredacted-MAX**\n4. Identity VL = **Qwen3-VL-4B-Instruct-Unredacted-MAX**, FP16, sdpa; 4-bit if OOM\n\nDo not use 8B/32B VL. Do not feed the still into the TEXT rewriter.\n"],"color":"#222","bgcolor":"#000"},{"id":19,"type":"PrimitiveStringMultiline","pos":[-842.9186789792956,158.94735054154341],"size":[520,320],"flags":{},"order":15,"mode":0,"inputs":[{"localized_name":"value","name":"value","type":"STRING","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"STRING","name":"STRING","type":"STRING","links":[14]}],"title":"USER IDEA (edit me)","properties":{"Node name for S&R":"PrimitiveStringMultiline","ue_properties":{"widget_ue_connectable":{},"version":"7.8","input_ue_unconnectable":{}}},"widgets_values":["A woman walks into a neon-lit hotel lobby at night, slow cinematic push-in, rain on the glass doors, reflections on marble floor, elegant coat, confident expression."],"color":"#323","bgcolor":"#535"}],"links":[[1,5,0,16,0,"*"],[2,6,0,16,1,"*"],[3,7,0,17,0,"*"],[4,8,0,17,1,"*"],[5,9,0,17,2,"*"],[6,10,0,17,3,"*"],[7,11,0,17,4,"*"],[8,12,0,17,5,"*"],[9,13,0,17,6,"*"],[10,14,0,17,7,"*"],[11,15,0,17,8,"*"],[12,16,0,18,0,"STRING"],[13,17,0,18,1,"STRING"],[14,19,0,32,0,"STRING"],[15,18,0,22,8,"STRING"],[16,22,0,23,0,"STRING"],[18,20,0,27,0,"IMAGE"],[19,27,0,29,0,"IMAGE"],[21,29,0,31,0,"*"],[22,30,0,31,1,"*"],[23,31,0,32,1,"STRING"],[24,32,0,22,6,"STRING"]],"groups":[{"id":1,"title":"CONTENT · SFW","bounding":[-2000,780,520,430],"color":"#3f789e","flags":{}},{"id":2,"title":"CONTENT · NSFW","bounding":[-1480,780,520,430],"color":"#a1309b","flags":{}},{"id":3,"title":"TARGET · MiniMax H3","bounding":[-2000,1260,560,430],"color":"#3f789e","flags":{}},{"id":4,"title":"TARGET · LTX-2.5","bounding":[-1440,1260,560,430],"color":"#a1309b","flags":{}},{"id":5,"title":"TARGET · Z-Image Turbo","bounding":[-880,1260,560,430],"color":"#a1309b","flags":{}},{"id":6,"title":"TARGET · Krea 2","bounding":[-2000,1680,560,430],"color":"#a1309b","flags":{}},{"id":7,"title":"TARGET · Qwen-Image","bounding":[-1440,1680,560,430],"color":"#a1309b","flags":{}},{"id":8,"title":"TARGET · Wan 2.2","bounding":[-880,1680,560,430],"color":"#a1309b","flags":{}},{"id":9,"title":"TARGET · Flux / Flux.2","bounding":[-2000,2100,560,430],"color":"#a1309b","flags":{}},{"id":10,"title":"TARGET · SDXL / Pony","bounding":[-1440,2100,560,430],"color":"#a1309b","flags":{}},{"id":11,"title":"TARGET · Universal Detail","bounding":[-880,2100,560,430],"color":"#a1309b","flags":{}},{"id":12,"title":"OPTIONAL · Ref Image","bounding":[-830,510,480,1000],"color":"#a1309b","flags":{}}],"config":{},"extra":{"ds":{"scale":1.0717944050000014,"offset":[1116.281027551997,17.07383912610726]},"info":{"name":"Prompt Creator — SFW / NSFW · model targets","description":"Local Qwen rewriter. Enable one CONTENT group and one TARGET group. Optional 4B VL identity pass at 768px for a reference still."},"ue_links":[]},"version":0.4}