Image generation

How it works

Write a prompt and hit Generate. The request goes to the relay, which hands it to an available render node, an independent GPU running a node-based pipeline built around the Aperture HD model. The finished image comes back to you. The render happens on a contributor's card, not on TALOS-owned hardware.

Privacy

Images are never stored on TALOS servers. Each render is returned to you and then dropped: no server-side copy, no public feed, no retained prompt history. Generations are saved only in your own browser and you can clear them whenever you like. Download an image to keep it; that's the only copy that lasts.

Content policy

The model is unfiltered, with one hard limit: no sexual content involving minors, checked against the prompt before anything renders. Adult content sits behind an 18+ toggle, off by default. With it off, an output classifier keeps adult imagery from coming back; with it on, nothing is scanned.

Using /create

• Prompt: describe the shot. Writing it like a real photo (candid photo, 35mm film, natural light) beats stacking quality words like "ultra detailed, 8k, sharp".
• Style presets: Photo (default), Cinematic, Anime, Digital Art, 3D. Photo is tuned for realism rather than the over-processed "AI" look.
• Aspect ratio: Square (1024×1024), Portrait (832×1216), Landscape (1216×832).
• NSFW (18+): toggle to allow adult content, with a one-time age confirmation.
• Advanced: negative prompt, steps, guidance (CFG) and seed.

Pricing

Flat 18 credits ($0.18) per image, drawn from your credit balance or your staker daily allowance. You're charged on success and refunded automatically if a render fails.

Model

Aperture HD: an open, unfiltered diffusion model. TALOS ships tuned defaults (a baseline anti-artifact negative prompt and real-photo styling) so a plain prompt comes out looking like a photograph instead of a render.

API

Image generation is also available over HTTP: see the API reference.