{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidh7b3fymqdltc7ov4ilk4dufopblnnk6bbuux64gafug6ikm4fb4",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mlzhtgqm2ay2"
  },
  "path": "/t/may-2026-chatgpt-api-image-gallery-prompt-tips-and-help-generative-art-theme-science/1378298?page=54#post_1097",
  "publishedAt": "2026-05-17T03:26:24.000Z",
  "site": "https://community.openai.com",
  "textContent": "This forum topic is a “like” factory for community visitors with low bar of entry.\n\nSince coding is not a hot topic, let’s go back to API prompting - using `gpt-image-2`, and a prompt to turn it into a set application.\n\n### Animation to real-life\n\n\n    You are a photographic image creator who reimagines artist fiction as real life: given cartoon/anime reference images, produce a high-fidelity, photoreal recreation of the characters and scene as if cast in a real-world movie adaptation or photographed as expertly staged cosplay. Select real people whose ethnic appearance, hair color, facial features, body type, and proportions best match the characters; reproduce clothing exactly with believable fabrics, seams, and tailoring in the same style and cut; and preserve the characters’ poses, gestures, body type, and overall composition so they remain immediately recognizable. Where the reference contains stylized or impossible details, correct them in physically plausible ways while retaining any exaggerated proportions that are essential to character identity and seen in input. Render skin, hair, and fabric textures with natural microdetail, realistic lighting and cast shadows, and accurate contact with the environment (footing, drape, reflections). Hair shall be in the natural range of people, unless the hair color is character-specific needing hair dye or a wig. For outdoor sunlight scenes, use believable directional sunlight, appropriate fill and reflector effects, realistic specular highlights, and consistent color temperature and dynamic range. Recreate the scene: For full body input, frame the subjects so all are shown head-to-toe and use the full output canvas with balanced composition; choose an appropriate focal length and depth of field for a full-body photograph (avoid extreme wide-angle distortion). If the character subject(s) are tightly framed purposefully in the input image, expand and zoom out the image only to where you avoid the need to fabricate unseen or unknown details out-of-frame. If no background exists in the reference, create a plausible, unobtrusive real-world setting consistent with the scene and lighting, using shallow depth or bokeh to keep focus on the subjects. Prioritize photographic realism: high resolution, natural skin tones, to produce a convincing, professional-quality digital photograph.\n\n    # User notes\n    *includes instructions and corrections needed from previous iterations*\n\n\nLet’s try some images that need a bit of thought.\n\nThere were several symptoms with vision on the edits endpoint, seen beyond what I show - sending the input either at original size or at a size vastly increased or matching a large output spec, details and the very nature of animated objects seem to be misinterpreted. Things we can see in drawings, the AI can’t see.\n\nSent:\n\nTune-ups required in the prompt’s allotted user input space, to get the variations deterministically reasonable:\n\n> Jane is Caucasian and has feminine proportions.\n>  Hosiery is sheer and not solid.\n>  Short pants and boots are dark grey, not the brown of the input image which has color grade issues.\n>  The dark device on the bottom of the easel is a handle for tightening the tripod.\n>  Jane’s art is also converted to real-life and painterly, not cartoonish. For example, the central picture is Jane’s reinterpretation of a Van Gogh.\n>  Observe the grass at the bottom: Jane is at an outdoor art show.\n\n* * *\n\nWell - we got overly “realistic”, I suppose.\n\nNeeding more prompt tune-ups by the user.\n\n> The small “angel” and “devil” versions of Daria in the image remain small, super-imposed on the photograph in post. They are each depicting the central character, but in the role of devil or angel. The photograph gives the impression they are actually existing in three dimensions and floating alongside the main character’s head, in the original poses.\n\nThe little characters are fun to look at, but the main character is unsettling. The direction that character eyes are looking is also very hard for the AI computer vision to pick out of anime images, needing instruction and intervention. Is the “chair” not reliably seen?\n\nThe output here is “quality”:“high” and a large image size by parameter, much larger than the forum displays before you click. $0.4037 per attempt by billed usage.",
  "title": "May 2026 — ChatGPT / API Image Gallery, Prompt Tips, and Help: Generative Art Theme: Science"
}