Back to Claude Scientific Skills

OpenRouter image model reference

skills/generate-image/references/models.md

2.57.06.3 KB
Original Source

OpenRouter image model reference

Snapshot of GET https://openrouter.ai/api/v1/images/models, taken 2026-07-26 (38 models). The catalogue moves; re-read it rather than trusting this table:

bash
python scripts/generate_image.py --list-models

No API key is needed to list models.

Capability table

refs is the maximum number of input_references accepted; n is images per request.

ModelnrefsstreamParameters
google/gemini-3.1-flash-image114noaspect_ratio, resolution, n, input_references
google/gemini-3.1-flash-image-preview114noaspect_ratio, resolution, n, input_references
google/gemini-3.1-flash-lite-image114noaspect_ratio, resolution, n, input_references
google/gemini-3-pro-image114noaspect_ratio, resolution, n, input_references
google/gemini-3-pro-image-preview114noaspect_ratio, resolution, n, input_references
google/gemini-2.5-flash-image13noaspect_ratio, n, input_references
black-forest-labs/flux.2-pro18noaspect_ratio, output_format, seed, n, input_references
black-forest-labs/flux.2-max18noaspect_ratio, output_format, seed, n, input_references
black-forest-labs/flux.2-flex18noaspect_ratio, output_format, seed, n, input_references
black-forest-labs/flux.2-klein-4b14noaspect_ratio, output_format, seed, n, input_references
bytedance-seed/seedream-4.51014noaspect_ratio, resolution, seed, n, input_references
openai/gpt-image-21016yesaspect_ratio, background, quality, output_compression, n, input_references
openai/gpt-image-11016yesaspect_ratio, background, quality, output_compression, n, input_references
openai/gpt-image-1-mini1016yesaspect_ratio, background, quality, output_compression, n, input_references
openai/gpt-5.4-image-21016yesbackground, quality, output_compression, n, input_references
openai/gpt-5-image1016yesbackground, quality, output_compression, n, input_references
openai/gpt-5-image-mini1016yesbackground, quality, output_compression, n, input_references
recraft/recraft-v4.1-pro61noaspect_ratio, n, input_references
recraft/recraft-v4.161noaspect_ratio, n, input_references
recraft/recraft-v4.1-vector61noaspect_ratio, n, input_references
recraft/recraft-v4.1-pro-vector61noaspect_ratio, n, input_references
recraft/recraft-v4.1-utility61noaspect_ratio, n, input_references
recraft/recraft-v4.1-utility-pro61noaspect_ratio, n, input_references
recraft/recraft-v4-pro61noaspect_ratio, n, input_references
recraft/recraft-v461noaspect_ratio, n, input_references
recraft/recraft-v4-vector61noaspect_ratio, n, input_references
recraft/recraft-v4-pro-vector61noaspect_ratio, n, input_references
recraft/recraft-v361noaspect_ratio, n, input_references
sourceful/riverflow-v2.5-pro110noaspect_ratio, background, output_format, resolution, n, input_references
sourceful/riverflow-v2.5-fast14noaspect_ratio, background, output_format, resolution, n, input_references
sourceful/riverflow-v2-pro110noaspect_ratio, resolution, n, input_references
sourceful/riverflow-v2-fast14noaspect_ratio, resolution, n, input_references
microsoft/mai-image-2.5-pro11noaspect_ratio, n, input_references
microsoft/mai-image-2.511noaspect_ratio, n, input_references
krea/krea-2-large1noaspect_ratio, resolution, seed, input_references
krea/krea-2-medium1noaspect_ratio, resolution, seed, input_references
krea/krea-2-medium-turbo1noaspect_ratio, resolution, seed, input_references
x-ai/grok-imagine-image-quality13noaspect_ratio, resolution, n, input_references

Reading the parameter columns

Parameters are not universal, and an unsupported one is rejected, not ignored:

  • resolution (512, 1K, 2K, 4K) — Gemini, Seedream, Riverflow, Krea, Grok. Not FLUX.
  • output_format — FLUX (png, jpeg) and Riverflow 2.5. Gemini does not take it.
  • quality, background, output_compression — the OpenAI family (background also on Riverflow 2.5).
  • seed — FLUX, Seedream, Krea. Not Gemini, not OpenAI.
  • aspect_ratio — nearly everything, but the enum differs. FLUX 2 Pro accepts 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, auto.

Per-model detail, including provider-specific passthrough parameters and pricing:

bash
curl -s https://openrouter.ai/api/v1/images/models/black-forest-labs/flux.2-pro/endpoints

FLUX 2 Pro, for example, allows steps, guidance, and safety_tolerance as passthrough parameters and bills $0.03 per output megapixel.

Choosing a model

  • General quality, prompt adherencegoogle/gemini-3.1-flash-image (skill default), or google/gemini-3-pro-image for the higher tier.
  • Photoreal and artistic control, seeded reproducibilityblack-forest-labs/flux.2-pro.
  • Cheap iterationblack-forest-labs/flux.2-klein-4b, google/gemini-3.1-flash-lite-image.
  • Batchesbytedance-seed/seedream-4.5 or the OpenAI family (up to 10 per request).
  • True vector output (SVG)recraft/recraft-v4-vector, recraft/recraft-v4.1-vector, and the -pro-vector variants. These return media_type: image/svg+xml.
  • Transparent backgrounds — the OpenAI family via --background transparent.
  • Heavy multi-reference compositing — OpenAI (16 references), Gemini and Seedream (14).

Provider routing

The Image API accepts the same provider block as chat completions — provider.only, provider.order, provider.ignore, provider.sort (price, throughput, latency), and provider.allow_fallbacks. The bundled script does not expose these; send a direct request when routing control matters.

Billing

Image billing is all-or-nothing: a generation either completes and is billed in full, or fails and is not billed. Partial preview frames delivered during streaming are not charged separately. Per-request cost comes back in usage.cost, which the script prints.