Anime to Real Pro IV · publiczny skill Open Skill
Anime to Real Pro IV
Anime to Real Pro IV to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/48h83m24.jpg
# Anime to Real Pro IV Turn an anime / manga / illustration image uploaded by the user into a hyper-realistic photo of the same character setup via **image-to-image**: facial features, hairstyle, outfit, pose, and overall composition all carry over from the original image — only the rendering style changes from 2D anime to real photography. ## When to use - The user uploads an anime / manga / 2D illustration and says "make it realistic", "anime to real Pro IV", "turn into a real person", "make it a photo", etc. - The user is unhappy with the previous realistic result and asks to re-run it, switch the style, or adjust age or skin tone. - Black-and-white manga, line art, and partially colored images also go through this flow (ask for natural colorization in the prompt). When judging the input: if the uploaded image is in a style other than anime/photography and the user wants a "shot-on-camera feel", use this skill. ## Steps ### 1. Confirm the input image - **A source image is required.** If the user asks for "anime to real" without uploading an image, ask them to upload the original first; **do not** substitute text-to-image generation. - Use that image as the **first** entry in `input_urls` (pass only this one): prefer an `asset_id` for assets already generated in this conversation; `s3://` or a real https URL also works. - Infer age, skin tone, shot framing, etc. from the original image yourself and start right away; only ask a follow-up question when the original truly cannot be judged (e.g., only a back view, or no person at all). ### 2. Assemble the English prompt (three parts) Image prompts are always in **English**; assemble them from the three parts below — none is optional: 1. **Realism side**: what style of final image you want. 2. **Consistency side**: the settings from the original image that must be preserved exactly. 3. **Composition-consistency side**: the top/bottom/left/right of the frame must not change. A template you can use directly (replace what's inside `<>` with the user's input): ``` Photorealistic ultra-detailed real-world photograph of the same character, <age, gender, complexion/skin tone, hair color and length, eye color>, identical hairstyle, identical outfit, identical accessories, identical facial features, identical pose and expression, natural skin texture with pores and fine details, realistic fabric and hair material, natural soft lighting, shot on 85mm lens, shallow depth of field, sharp focus, 2K high definition. Keep the original composition exactly: same camera angle, same subject position, same subject size, same background layout, same left/right/top/bottom crop — do not reframe, do not zoom in or out, do not extend or pad the canvas, do not add or remove any element. If the input is line art or monochrome manga, add natural colorization while keeping the original design. ``` Key points: - Age and skin tone **must be written into the prompt** (these are the two things realism conversion most easily gets wrong). - Explicitly state "the same setup": facial features, hairstyle, accessories, outfit style and color scheme all unchanged. - Hard-code composition consistency with the English phrasing "same camera angle / same position / same size / no reframe / no extend"; this is a hard constraint of this skill. - If the user adds extra text requirements (hair color, scene, lighting, mood), fold them into the realism side; **do not** change the composition sentence because of them. ### 3. Choose the channel and parameters - `model` may only be chosen from the three image-to-image channels below and **must be passed explicitly**; physical model names are forbidden: - `i2i.hd.lite` — the default first choice; fast, try it first. - `i2i.hd` — use when higher detail and fidelity are needed (user says "sharper" or "more realistic"). - `i2i.hd.gpt` — use when the user gives longer or very specific text requirements and stronger instruction-following is needed. - `input_urls=[or