Photoreal Stills

Make images that look like real photographs. The prompt does most of the work: subject, action, place, light, and camera, in that order.

Inputs to collect

• Subject and action: who or what, doing what.
• Place and time: setting, time of day, weather, season.
• Look: documentary, editorial, stock, cinematic, or a film look.
• Format: aspect ratio or pixel size per use (16:9 hero, 4:5 feed, 1:1 card), and image count.
• Limits: no real, named people without consent; no brand logos the user does not own.

Models

<!-- tokenlab:models:start --> <!-- Generated from the live catalog by scripts/sync_models.py. Do not edit; change preferredModels in catalog.json. -->

Operation	Model	Pick	Use
text-to-image	flux-2-pro	default	General photoreal
text-to-image	flux-2-max	alternative	Highest-detail hero image (not live-tested)
text-to-image	seedream-5.0-pro	alternative	Photoreal, alternative look (not live-tested)
text-to-image	krea-2-large	alternative	Photoreal, alternative look (not live-tested)
text-to-image	stable-image-ultra	alternative	Photoreal, alternative look (not live-tested)
text-to-image	flux-2-klein-9b	alternative	Low-cost draft

This list refreshes daily from the live catalog. Preferred models come first; otherwise models are ranked by their 24-hour success rate. Before a call, confirm the model with GET /v1/models/{model} and read its price from GET /v1/models/{model}/pricing.

<!-- tokenlab:models:end -->

How to choose:

• General photoreal: FLUX.2. Its top tier gives the most detail for a hero image.
• Alternative look: Seedream, Krea, or Stable Image Ultra.
• Low-cost draft: a FLUX.2 Klein model.

All use /v1/images/generations. Send operation: "text-to-image" when the contract lists operation (Stable Image Ultra lists only prompt and aspect_ratio). Exact fields come from each model's contract.

Workflow

Run each call with tokenlab-run (contract → estimate → call → poll → result). Commands below use its helper, ../tokenlab-run/scripts/tokenlab_run.py, a path relative to this skill's folder. If you installed this skill without tokenlab-run, send the same requests with curl. --wait stops after 100 s; a longer task keeps running, so continue with the resume command it prints.

1. Read the contract. Note the size field: resolution as WIDTHxHEIGHT, aspect_ratio, width / height, or none. Send only listed fields.
2. Estimate the cost from GET /v1/models/{model}/pricing: price per image times the image count.
3. Write the prompt (see Technique). Set the frame size for the use.
4. Read the response. A sync response has data[].url. A response with status: "pending" is a task: poll GET /v1/tasks/{id} every 5–10 s until completed or failed. In tests, a FLUX.2 model returned a task without async: true.
5. Check the result against the Quality bar. Re-run with a changed prompt, not the same one, when a detail is wrong.
6. Return the result URLs, the model, the task IDs, and the prompt you sent.

Technique

• Order the prompt by priority. Black Forest Labs' FLUX.2 guide: "Subject + Action + Style + Context", and word order matters because the model pays more attention to what comes first.
• Aim for 30–80 words. The FLUX.2 guide calls this range usually ideal. Go longer only for complex scenes.
• Name the camera, lens, and light. "Shot on Fujifilm X-T5, 35mm f/1.4" gives a more real result than "professional photo" (FLUX.2 guide). Add the light: "warm tungsten stall lights", "overcast soft daylight", "low sun from behind".
• Use positive words, not negatives. FLUX.2 does not support negative prompts. Say "sharp focus throughout", not "no blur".
• Colors by hex. For brand colors on FLUX.2, tie the code to an object: "the awning is color #1F6F5C".
• Skin and hands. Ask for "natural skin texture" and keep hands in a simple pose, such as holding one object. Check fingers in every result.
• Frame size. On the FLUX family, resolution takes pixels: a FLUX.2 Klein model with 1344x768 returned 1344 × 768. A wrong value returns 400 with param: "resolution". P-Image lists width / height, but a 768 × 768 test returned 1024 × 1024; use it for square drafts.
• Draft cheap, then finish. Explore the composition with a low-cost model, then run the chosen prompt on the default for the final image.

Quality bar

• Looks like a photo: real light falloff, believable depth of field, no plastic skin, no painted look.
• Hands, teeth, eyes, and text-like shapes in the background have no obvious errors.
• Light direction and color are the same on the subject and the background.
• The frame size and aspect ratio fit the use.
• No real person's likeness or brand mark that the user did not ask for.

Related skills

• tokenlab-run: the call, poll, and result mechanics.
• tokenlab-product-photography: a product as the subject.
• tokenlab-text-in-image: readable words in the image.
• tokenlab-character-consistency: the same person across a series.
• tokenlab-edit-image: change an existing photo.
• tokenlab-prompting: prompt craft per model family.