Bria API vs Iris-3B

Bria API

Per-image prices, 100 free calls without a card, OpenAPI files and Markdown docs make a first call quick, and the terms warrant that customer content is not used for training. Keys are organisation-wide with no scopes, and the API has no changelog or deprecation policy.

Iris-3B

Code and weights are Apache-2.0, and one Python package runs text-to-image, a 4-step Turbo, depth and 4x restoration from a documented CLI. It is ten days old, with no PyPI package, version tags, changelog or CI, needs an NVIDIA GPU and about 12 GB of weights, and has no server or API of its own.

Bria API

1. Send the key in the api_token header, not as Authorization: Bearer. Only /v2/image/verify and /v2/video/verify work without it
2. Expect 202 with request_id and status_url from v2 endpoints. Poll GET /v2/status/{request_id} every 1 to 2 seconds, or pass sync: true
3. Status polls count towards the per-endpoint limit (10 a minute on Free, 60 on Pay-as-you-go), so prefer webhook_url for batches
4. Download results promptly. Image URLs expire after 3 days by default, and legacy /v1/product/* results after 1 hour
5. Do not resend a failed POST blindly. There is no idempotency key, and the SDKs retry status polls only

Iris-3B

1. Check for an NVIDIA GPU with CUDA and Python 3.11 or newer before downloading. The base weights are about 12 GB, and Qwen3-VL-4B downloads on the first run
2. Exclude depth/*, upscaler/* and turbo/* in hf download unless those models are needed. Each of the first two adds about 12 GB
3. Don't pass --steps, --cfg-scale, --negative-prompt, --order or --shift with a Turbo checkpoint. The script exits with an error
4. Pass --seed to reproduce an image, and set --outdir, since files are named after the prompt and a repeat run overwrites them
5. Load only exported directories (model.safetensors and config.yaml). A training .pth is read with torch.load(weights_only=False), so never load one from an untrusted source