AI Image Generation APIs Cheat Sheet
This cheat sheet covers common patterns for calling image generation APIs — model IDs, request parameters, prompting syntax, and cost math.
Getting Started
Image models {.row-span-2}
| Model | Known for | API model ID |
| FLUX 1.1 Pro | Photorealism, prompt adherence | flux-pro/v1.1 |
| FLUX Schnell | Speed, drafts | flux/schnell |
| Stable Diffusion 3.5 | Styles, open weights | stable-diffusion-v35-large |
| DALL·E 3 | Instruction following | dall-e-3 |
| Imagen 4 | Clean detail | google/imagen-4.0-generate-001 |
| Recraft V3 | Brand styles, SVG | recraft-v3 |
{.show-header}
IDs shown for a multi-provider endpoint; direct provider APIs use their own naming.
Basic call
curl https://api.aimlapi.com/v1/images/generations \
-H “Authorization: Bearer $API_KEY” \
-H “Content-Type: application/json” \
-d ‘{
“model”: “flux-pro/v1.1”,
“prompt”: “isometric illustration of a data center, soft colors”,
“size”: “1024×1024”
}’
Image APIs are synchronous: one request, one response.
Basic call (Python)
import requests
resp = requests.post(
“https://api.aimlapi.com/v1/images/generations”,
headers={“Authorization”: f”Bearer {API_KEY}”},
json={
“model”: “flux-pro/v1.1”,
“prompt”: “isometric illustration of a data center”,
“size”: “1024×1024”,
},
)
url = resp.json()[“data”][0][“url”]
Response handling
- response_format: url (default) or b64_json
- Result URLs usually expire (minutes–hours) — download and store in your own bucket
- Failed generations are often still billed — validate prompts before batch runs
- Same prompt + seed = same output; cache instead of regenerating
Parameters and Prompting
Common parameters {.col-span-2}
| Parameter | Aliases | What it does |
| size | width + height, aspect_ratio | Output dimensions; models allow fixed sets |
| n | num_images | Batch count per request |
| seed | — | Same seed + prompt ≈ same image |
| steps | num_inference_steps | Quality vs speed (diffusion models) |
| guidance | cfg_scale | Prompt adherence; too high = artifacts |
| negative_prompt | — | What to avoid (SD family; some models ignore) |
| response_format | — | url (default, expires) or b64_json |
{.show-header}
Unsupported parameters are silently ignored — check the response, not the docs.
Prompt structure
[subject] [action], [environment], [style],
[lighting], [camera]
portrait of a violinist mid-performance,
concert hall bokeh, editorial photography,
dramatic rim lighting, 85mm f/1.4
# Weighting (SD-family syntax)
(golden hour:1.3), (crowd:0.6)
Prompt add-ons
| Goal | Add |
| Photorealism | 35mm, f/2.8, RAW photo |
| Illustration | flat vector, isometric |
| Brand consistency | fix seed + reuse style block |
| Clean composition | generous whitespace, centered |
{.show-header}
Reference
Cost ballpark
| Media | Range |
| Image, standard 1024² | $0.02–0.08 |
| Image, premium models | $0.10–0.25 |
{.show-header}
- Published list prices at time of writing — verify with your provider
- Higher resolutions and quality tiers multiply the price
Common errors
| Problem | Fix |
| 400 invalid size | Models allow a fixed dimension list — check catalog |
| Content policy rejection | Rephrase; avoid real names, brands, violence |
| 429 | Image rate limits are lower than LLM limits — queue jobs |
| Expired result URL | Download on completion, store yourself |
{.show-header}
Video APIs
Same auth and endpoint family, but asynchronous: submit a job, poll for status, download the result. Check your provider’s video docs for the job contract.
Also see
- Black Forest Labs docs _(FLUX parameters)_
- AI/ML API docs _(endpoint and model IDs used in the examples above)_