PixArt-Σ
Free.ai (self-hosted)
·
image
·
~1000 tokens per image
·
2.8
from 106 users of this category
PixArt-Σ is an image generation model built by PixArt-alpha. Strongest at Ultra-high-resolution generation on modest GPUs.. Self-hosted on Free.ai GPUs — runs free against your daily token pool (1,000 tokens per image). Released under OpenRAIL++ — commercial use permitted on Free.ai.
Similar models
Frequently Asked Questions
PixArt-Σ is an image generation model built by PixArt-alpha. Strongest at Ultra-high-resolution generation on modest GPUs.. Self-hosted on Free.ai GPUs — runs free against your daily token pool (1,000 tokens per image). Released under OpenRAIL++ — commercial use permitted on Free
PixArt-Σ works well for Ultra-high-resolution generation on modest GPUs.. Try the sample prompts above to see its strengths.
About 1,000 tokens per image, shown live before you generate. A free Free.ai account includes a 30,000-token daily pool, so most users can try PixArt-Σ without spending anything; beyond it, token top-ups start at $1.
It depends on the prompt. Free.ai lets you compare PixArt-Σ side-by-side with any other model at /image/compare/ - pick a sibling from the alternatives list to see how they differ on the same input.
Yes. Free.ai does not add watermarks and the outputs are yours to use commercially. The underlying model is OpenRAIL++-licensed. See /models/pixart-sigma/ for the full breakdown.
On Free.ai, PixArt-Σ supports 1:1 (square), 16:9 (landscape), 9:16 (portrait), 4:3, and 3:4. Pick the ratio in the form above before generating.
Most images come back in 5-25 seconds depending on the model and current GPU load. The result appears inline; signed-in users can also Add to queue to close the tab and get notified when it lands.
Yes - pick 1, 2, or 4 in the count field. Each image deducts about 1,000 tokens. For larger batches use /image/generate/ which fans out the same prompt across multiple models in parallel.
Free.ai does not train models on your prompts. Self-hosted models stay on our GPUs. Premium models route to the upstream provider for inference. Either way we never resell your data.
Yes. POST to /v1/image/generate/ with model="pixart-sigma" and include your Bearer token. Full reference: /api/.
PixArt-Σ is OpenRAIL++-licensed with 600M parameters - see /models/pixart-sigma/ for the model card and our open-source repos at github.com/freeaigit for ready-to-run scripts.
Free accounts get a 30,000-token daily pool. When that runs out, token top-ups start at $1, pay-as-you-go - no subscription required.