agent.md

# X Thread To Telegram Draft - AIWarper - 2026-06-03

## Source

- Main tweet: `https://x.com/AIWarper/status/2062239306140590164`

## Pipeline

1. Generate reference images with GPT-Image-2.
2. Use the first reference as a strict identity anchor for the blonde woman and place her naturally in a five-person living-room scene.
3. Generate a second reference: a ground-level shot following a white cat entering the same scene.
4. Draw a simple red path over the image. The author explicitly treats this as a rough direction doodle, not a technical diagram.
5. Seedance, attach the red-path sketch and the cat reference frame.
6. Generate a 1080p, 10-second, 16:9 video as a continuous third-person follow shot.

## Rebuilt Prompt Pack

These are cleaned, Telegram-ready prompts based on the thread. They preserve the original English direction without copying the full thread verbatim.

### GPT-Image-2 Prompt A

Use the first reference image as a strict identity reference for the blonde woman. Create a photorealistic 16:9 scene of five alternative/trendy women hanging out in a warm, lived-in rented-house living room. Preserve the blonde woman's face, blue eyes, hair, and outfit identity, and place her naturally on the sofa. Add four distinct women with contrasting alt/Y2K/grunge styling. Make the room cozy and personal: sofa, rug, plants, framed art, band posters, warm lamps, books, shelf clutter, candid documentary energy, natural skin/fabric texture, and relaxed group-hangout composition.

### GPT-Image-2 Prompt B

Create a ground-level shot from behind a white cat as it walks into the same living-room scene. Keep the environment, lighting, camera language, and visual identity consistent with the previous reference.

### Path Direction Step

Draw a red route line over the reference image showing exactly where the cat should move. Keep it rough and readable. The path is a motion guide, not a final visual element.

### Seedance Prompt

Third-person continuous follow shot behind the white cat. The cat follows the red path from the provided guide image and jumps onto people according to that route. Keep the red path and arrows invisible in the final output. No cuts, no jump cuts, one continuous take.

Settings Example:

- `1080p`
- `10s`
- `16:9`

## Telegram Post Draft

AI video workflow worth saving: use a rough red-line doodle as motion direction, not just a written prompt.

The result: a white cat moves through a crowded living-room scene, following a path that was drawn directly over the reference frame. The author says this was the first output, without cherry-picking.

Pipeline:

1. Generate two reference images with GPT-Image-2.
2. Build the first image around a strict character identity reference.
3. Generate a second frame from a ground-level cat-following angle.
4. Draw a simple red path over the frame.
5. Use an LLM to analyze the path sketch and reference frame, then generate a detailed Seedance prompt that converts the drawn route into camera and character motion instructions.
6. Send the path sketch + cat frame into Seedance.

The key idea is simple: when motion is hard to describe, draw it. A rough path can carry more direction than another paragraph of prompt text.

Prompt structure:

```text
Reference image: strict identity anchor + target scene.
Second image: ground-level cat POV entering the same scene.
Path image: rough red motion route.
Video prompt: continuous third-person follow shot, obey the red route, hide the guide line, no cuts.
```

Tools mentioned:

- GPT-Image-2 for references
- Seedance for video

Source: `https://x.com/AIWarper/status/2062239306140590164`