backend new models check - replicate - 2026-09-16

Files checked: `src/backend/drivers/ai-image/providers/replicate/models.ts`

Backend's replicate image list is very small (only 8 models: `flux-2-pro`, `flux-2-dev`, `flux-2-klein-9b-base`, `flux-2-klein-4b`, `flux-schnell`, `flux-1.1-pro`, `leonardoai/lucid-origin`, `leonardoai/phoenix-1.0`) against Replicate's ~440-model "official models" collection (not a full catalog — no task field, so image-generation-ness was judged by name/description). Utility-only tools (background removal, pure upscalers, colorization, vectorize-only, OCR, moderation/classification, LoRA trainers/fine-tune bases) and anything video/audio/3D/embedding were excluded as out of scope.

**image (143)** — genuine image generation/editing models not in the backend:

`black-forest-labs/flux-1.1-pro-ultra`, `black-forest-labs/flux-2-flex`, `black-forest-labs/flux-2-klein-4b-base` (distinct from backend's `flux-2-klein-4b`), `black-forest-labs/flux-2-klein-9b` (see note below), `black-forest-labs/flux-2-max`, `black-forest-labs/flux-canny-dev`, `black-forest-labs/flux-canny-pro`, `black-forest-labs/flux-depth-dev`, `black-forest-labs/flux-depth-pro`, `black-forest-labs/flux-dev`, `black-forest-labs/flux-fill-dev`, `black-forest-labs/flux-fill-pro`, `black-forest-labs/flux-kontext-dev`, `black-forest-labs/flux-kontext-max`, `black-forest-labs/flux-kontext-pro`, `black-forest-labs/flux-krea-dev`, `black-forest-labs/flux-pro`, `black-forest-labs/flux-redux-dev` (modality?), `black-forest-labs/flux-redux-schnell` (modality?), `bria/expand-image`, `bria/fibo`, `bria/fibo-edit`, `bria/generate-background`, `bria/genfill`, `bria/image-3.2`, `bria/product-packshot` (modality?), `bytedance/dreamina-3.1`, `bytedance/seedream-3`, `bytedance/seedream-4`, `bytedance/seedream-4.5`, `bytedance/seedream-5-lite`, `bytedance/seedream-5-pro`, `easel/ai-avatars` (modality?), `flux-kontext-apps/cartoonify`, `flux-kontext-apps/change-haircut`, `flux-kontext-apps/depth-of-field`, `flux-kontext-apps/face-to-many-kontext`, `flux-kontext-apps/filters`, `flux-kontext-apps/iconic-locations`, `flux-kontext-apps/impossible-scenarios`, `flux-kontext-apps/multi-image-kontext-max`, `flux-kontext-apps/multi-image-kontext-pro`, `flux-kontext-apps/multi-image-list`, `flux-kontext-apps/portrait-series`, `flux-kontext-apps/professional-headshot`, `flux-kontext-apps/restore-image`, `flux-kontext-apps/restyle-video-frame` (modality?), `flux-kontext-apps/text-removal`, `google/gemini-2.5-flash-image`, `google/imagen-3`, `google/imagen-3-fast`, `google/imagen-4`, `google/imagen-4-fast`, `google/imagen-4-ultra`, `google/nano-banana`, `google/nano-banana-2`, `google/nano-banana-2-lite`, `google/nano-banana-pro`, `ideogram-ai/ideogram-character`, `ideogram-ai/ideogram-v2`, `ideogram-ai/ideogram-v2-turbo`, `ideogram-ai/ideogram-v2a`, `ideogram-ai/ideogram-v2a-turbo`, `ideogram-ai/ideogram-v3-balanced`, `ideogram-ai/ideogram-v3-quality`, `ideogram-ai/ideogram-v3-turbo`, `ideogram-ai/ideogram-v4-balanced`, `ideogram-ai/ideogram-v4-quality`, `krea/krea-2-large`, `krea/krea-2-medium`, `luma/photon`, `luma/photon-flash`, `luma/reframe-image` (modality?), `minimax/image-01`, `openai/dall-e-2`, `openai/dall-e-3`, `openai/gpt-image-1`, `openai/gpt-image-1-mini`, `openai/gpt-image-1.5`, `openai/gpt-image-2`, `openai/gpt-image-2.5-flare`, `openai/gpt-image-2.5-sunburst`, `prunaai/flux-fast`, `prunaai/flux-kontext-fast`, `prunaai/hidream-l1-fast` (distinct family from together's HiDream-I1), `prunaai/p-image`, `prunaai/p-image-edit`, `prunaai/p-image-ideogram` (also new for together, separate provider entry), `prunaai/p-image-try-on`, `prunaai/wan-2.2-image`, `prunaai/z-image-turbo`, `qwen-edit-apps/qwen-image-edit-plus-lora-fusion`, `qwen-edit-apps/qwen-image-edit-plus-lora-next-scene`, `qwen-edit-apps/qwen-image-edit-plus-lora-photo-to-anime`, `qwen-edit-apps/qwen-image-edit-plus-lora-relight`, `qwen-edit-apps/qwen-image-edit-plus-lora-skin`, `qwen/qwen-edit-multiangle`, `qwen/qwen-image`, `qwen/qwen-image-2`, `qwen/qwen-image-2-pro`, `qwen/qwen-image-2512`, `qwen/qwen-image-edit`, `qwen/qwen-image-edit-2511`, `qwen/qwen-image-edit-plus`, `quiverai/arrow-1.1`, `quiverai/arrow-1.1-max`, `recraft-ai/recraft-20b`, `recraft-ai/recraft-20b-svg`, `recraft-ai/recraft-v3`, `recraft-ai/recraft-v3-svg`, `recraft-ai/recraft-v4`, `recraft-ai/recraft-v4-pro`, `recraft-ai/recraft-v4-pro-svg`, `recraft-ai/recraft-v4-svg`, `recraft-ai/recraft-v4.1`, `recraft-ai/recraft-v4.1-pro`, `recraft-ai/recraft-v4.1-pro-svg`, `recraft-ai/recraft-v4.1-svg`, `recraft-ai/recraft-v4.1-utility`, `recraft-ai/recraft-v4.1-utility-pro`, `retro-diffusion/rd-animation` (modality?), `retro-diffusion/rd-fast`, `retro-diffusion/rd-plus`, `retro-diffusion/rd-tile`, `runwayml/gen4-image`, `runwayml/gen4-image-turbo`, `sourceful/riverflow-2.0-fast`, `sourceful/riverflow-2.0-pro`, `sourceful/riverflow-2.0-refsr`, `sourceful/riverflow-v2.5-pro`, `stability-ai/stable-diffusion-3`, `stability-ai/stable-diffusion-3.5-large`, `stability-ai/stable-diffusion-3.5-large-turbo`, `stability-ai/stable-diffusion-3.5-medium`, `tencent/hunyuan-image-2.1`, `tencent/hunyuan-image-3`, `wan-video/wan-2.2-5b-fast` (modality?, description mixes t2i and i2v), `wan-video/wan-2.7-image`, `wan-video/wan-2.7-image-pro`, `wavespeedai/qwen-image`, `xai/grok-imagine-image`, `xai/grok-imagine-image-2`, `xai/grok-imagine-image-quality`

Note on `black-forest-labs/flux-2-klein-9b`: backend's `flux-2-klein-9b-base` entry carries a bare alias "flux-2-klein-9b" that actually appears to refer to the -base (undistilled) model, while this upstream id is the separate 4-step-distilled variant — worth verifying whether the backend's alias is a naming mix-up or intentional.

Excluded as out-of-scope/noise: fine-tune-base and LoRA-loader variants (`flux-*-finetuned`, `flux-*-lora`, `flux-2-klein-*-base-lora`, `p-image-lora`/`p-image-edit-lora`/`p-image-trainer`, generic `qwen-image-edit-plus-lora` loader, `qwen-image-lora-trainer-legacy`); pure utility tools with no generative content (upscalers, background removal/cutout/shadow, colorization, vectorize-only, OCR, NSFW classification, dust-and-scratch); and all video/audio/3D/chat/embedding entries.

legacy upstream models not in backend (verify intent): none flagged — replicate's collection carries no reliable timestamp signal to judge staleness.

Start with src/backend/drivers/ai-image/providers/replicate/models.ts and compare its current eight entries with the proposed Replicate model list in the issue. Verify ambiguous aliases and modality exclusions before deciding which entries belong. Done means the backend's Replicate model definitions reflect the verified in-scope image models without adding the listed out-of-scope tools or non-image models.

Written by the indexing model from the issue text.