Supported providers

DALL-E (OpenAI), Imagen (Google), Stable Diffusion XL family, Flux family (open weights), Recraft, Ideogram, plus our internal compute-fund hosted models for batch jobs.

Pricing dimensions

Image generation is billed PerImage with dimensional pricing tiers (e.g., 1024×1024 vs 1792×1024). PricingUnit and ImageDimensions are recorded on every BillingRecord so audits are reproducible. Quality multipliers (Standard / HD / Ultra) are first-class fields, not hidden in opaque cost.

Editing

ImageEditJob (inpainting, outpainting, variation, background removal) uses a separate code path and may use a different model than generation. Edit jobs carry a source image reference plus a mask or prompt; output is stored on Bunny CDN by default.

Safety

Every output passes through moderation before being shown to you. NSFW content is blocked at the Free tier and gated by jurisdiction-aware controls on paid tiers. CSAM detection runs at every tier and cannot be disabled.

Tips

• Use 1024×1024 for fast iteration, switch to 1792×1024 or 1024×1792 for final output.
• Negative prompts work best with SDXL and Flux; less effective with closed models.
• For brand-consistent output, fine-tune via Settings → Fine-Tune (Pro+) — this stays on-platform per ADR-031, no leakage to third-party training buckets.