AI DEPTH MAP GENERATOR · публичный навык Open Skill

AI DEPTH MAP GENERATOR

AI DEPTH MAP GENERATOR — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.

Other

8 3 9

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1463189/2026-09-13/basbatey.png

# Depth Map Generator

Turns an uploaded image into a depth-map style visualization (z-depth look) using the image
generation tool. NOTE: this is an AI-rendered depth-style pass, not true metric depth computed by
a depth-estimation model — it is ideal for parallax / ControlNet-style / depth-of-field / fog / VFX
looks, but not for precise measurement or stereo disparity.

## When to use
- User uploads an image and asks for a depth map, depth pass, z-depth, depth estimation, or
  parallax / 2.5D / 3D-photo / ControlNet-depth / fog / DoF prep.
- User only says "make it a depth map" without attaching a source image → ask for the image first.

## Steps
1. Take the uploaded image's `asset_id` / `uri` as the source (`input_urls`). Use its native
   orientation if known, else `square`.
2. Pick options from the request (default otherwise — ask only if truly ambiguous):
   - Style: `grayscale` (default) | `inferno` | `viridis`
   - Near polarity: `near bright` (default, nearest = white) | `near dark` (nearest = black)
   - `side-by-side` if the user wants the original next to the depth map.
3. Call `generate_image`:
   - `model='i2i.hd'` (fall back to bare `i2i` if `i2i.hd` is not available)
   - `input_urls=[<source image asset id/uri>]`
   - `orientation` = source's orientation (`landscape` / `portrait` / `square`)
   - `prompt` (adjust `[near]` = white or black, and `[style]` = "flat monotone grayscale" /
     "inferno color map" / "viridis color map"):
     > Convert this image into a clean single-channel depth map (z-depth) visualization. Nearest
     > surfaces appear [white / black], farthest surfaces appear [black / white], with a smooth
     > continuous depth gradient. No texture, no lighting, no recognizable detail — keep only the
     > subject layout and framing. [style]. Flat matte depth render, full frame.
   - For `side-by-side`: add to the prompt "split the frame: original on the left half, its depth
     map on the right half."
   - `name` = `{input_name}_depth_map`, `entity_type` = `depth_maps`.
4. Reply in one or two lines: what was produced, the style / near polarity used, and offer to flip
   either option if the result isn't what they need. Don't dump URLs.

## Video
True per-frame video depth is not supported by the generation tools. If the user wants a moving 3D /
parallax feel from a video, offer: grab a key still → depth-map it → animate the parallax via
`generate_video` (`i2v`). Do not claim per-frame video depth is possible.

## Notes
- Be honest with the user: the output is an AI-rendered depth-style interpretation, not metric
  distance, so it won't produce physically accurate depth values.
- If a result comes back colored or looks like a normal photo, retry with stronger "flat monochrome
  single-channel depth map, no lighting, no texture, no reflections" phrasing.

Поделиться Digen

AI DEPTH MAP GENERATOR · публичный навык Open Skill

AI DEPTH MAP GENERATOR