--- url: /orchestration/recipes.md --- > Documentation index: https://developer.civitai.com/llms.txt — every page on this site has a markdown version at its URL + `.md`. # Recipes Task-oriented, end-to-end examples. Each recipe walks through a real workflow: what to send, what you get back, common parameter tweaks, and troubleshooting. ::: tip Not the same as App Blocks recipes These are **Orchestration REST API** recipes — you hold a Bearer token and POST workflows to `orchestration.civitai.com` yourself. Not to be confused with **App Blocks [`customComfy` recipes](https://developer.civitai.com/apps/guide/comfy-cloud.md)**, which are server-registered ComfyUI workflows a sandboxed *block* invokes by id through the host bridge (the block never calls the orchestrator directly). ::: ## Video * [WAN video generation](https://developer.civitai.com/orchestration/recipes/wan.md) — all WAN versions (2.1–3.0) across FAL, Comfy, Civitai, and Alibaba, with text-to-video, image-to-video, reference-to-video, and edit-video operations * [LTX2 video generation](https://developer.civitai.com/orchestration/recipes/ltx2.md) — Lightricks LTX2 and LTX2.3 on Comfy, including the new videoToVideo (style transfer) and audioToVideo (talking-head) operations * [Kling video generation](https://developer.civitai.com/orchestration/recipes/kling.md) — Kuaishou Kling (v1/v1.5/v1.6/v2/v2.5-turbo with camera control) and Kling V3 (5 operations, multi-prompt, audio, video-to-video) * [Vidu video generation](https://developer.civitai.com/orchestration/recipes/vidu.md) — Vidu 2.0 (flat 600 Buzz, anime style, first-last-frame), Vidu Q3 (per-second pricing, 4 resolution tiers, turbo mode, native audio) and Vidu Q4 via FAL (image- and reference-to-video, up to 4K) * [Veo 3 video generation](https://developer.civitai.com/orchestration/recipes/veo3.md) — Google Veo 3.0/3.1 in standard / fast / lite tiers; operation inferred from image count; optional synchronized audio track * [Gemini Omni video generation](https://developer.civitai.com/orchestration/recipes/gemini-omni.md) — Google Gemini Omni 1.1 Flash; 360p–4K with audio, up to 6 reference images; billed on the seconds actually delivered * [MiniMax H3 video generation](https://developer.civitai.com/orchestration/recipes/minimax-h3.md) — MiniMax's native H3 model at 2K; text-to-video, first/last-frame, and reference-to-video with up to 9 reference images plus 3 videos and 3 audio clips * [MiniMax H3 (ComfyUI) video generation](https://developer.civitai.com/orchestration/recipes/minimax-h3-comfy.md) — the open H3 weights on Civitai's ComfyUI workers, up to 1344 px; text/frame-to-video, reference images, control video, LoRAs, turbo * [Seedance video generation](https://developer.civitai.com/orchestration/recipes/seedance.md) — ByteDance Seedance direct on BytePlus; v2 / v2-fast / v2-mini up to 15 s and 1080p, or v2.5 for 30 s single-shot and a 50-asset reference budget * [Grok video generation](https://developer.civitai.com/orchestration/recipes/grok-video.md) — xAI Grok-Imagine-Video via FAL; text-to-video, image-to-video, and edit-video with 480p/720p output * [FLUX-3 video generation](https://developer.civitai.com/orchestration/recipes/flux3-video.md) — Black Forest Labs FLUX-3 via FAL; text / image / first-last-frame / keyframes / extend plus draftEnhance, 720p–1080p with synchronized audio and a cheap draft-then-enhance loop * [Happy-Horse video generation](https://developer.civitai.com/orchestration/recipes/happy-horse.md) — Alibaba Happy-Horse via FAL; text-to-video, image-to-video, video editing, and up to 9 character references * [HunyuanVideo generation](https://developer.civitai.com/orchestration/recipes/hunyuan.md) — Tencent HunyuanVideo on Comfy workers; text-to-video with LoRA support; compute-intensive, always use `wait=0` * [Video upscaling](https://developer.civitai.com/orchestration/recipes/video-upscaler.md) — FlashVSR, 2–4× with a 2560 px output cap * [Video frame interpolation](https://developer.civitai.com/orchestration/recipes/video-interpolation.md) — VFIMamba, 2× or 3× frame-count, smooths generated or low-FPS footage * [Compose media (video)](https://developer.civitai.com/orchestration/recipes/compose-media-video.md) — overlay/stack videos on a canvas, place audio over a clip, picture-in-picture; the video form of `composeMedia` ## Image * [Flux 2 image generation](https://developer.civitai.com/orchestration/recipes/flux2.md) — Flux.2 Klein (default, cheap + capable, 4b/9b, supports createVariant) plus Dev / Flex / Pro / Max for higher-fidelity and commercial tiers * [Flux 1 image generation](https://developer.civitai.com/orchestration/recipes/flux1.md) — Flux.1 through sdcpp (default, minimal required input) or Comfy, plus the BFL-hosted `flux1-kontext` editing tier * [Z-Image generation](https://developer.civitai.com/orchestration/recipes/zimage.md) — lightweight text-to-image on sdcpp; `turbo` (default, distilled, extremely fast + cheap) or `base` when you need more fidelity * [Qwen image generation](https://developer.civitai.com/orchestration/recipes/qwen.md) — Qwen-Image 20B on sdcpp (default), the Alibaba-hosted catalog (`engine: "qwen"`), or Qwen2 under a FAL-shaped request with a Pro tier; supports createImage + createVariant + editImage * [MAI Image 2.5 image generation](https://developer.civitai.com/orchestration/recipes/mai-image.md) — Microsoft MAI Image 2.5 via FAL; text-to-image only, eleven aspect ratios (incl. `auto`), flat per-image pricing * [Anima image generation](https://developer.civitai.com/orchestration/recipes/anima.md) — anime-tuned sdcpp ecosystem with built-in diffuser, LoRA support, createImage only * [ERNIE image generation](https://developer.civitai.com/orchestration/recipes/ernie.md) — Baidu ERNIE Image on Comfy; `ernie` standard + `turbo` distilled variant, built-in diffuser, LoRA support, createImage only * [SDXL image generation](https://developer.civitai.com/orchestration/recipes/sdxl.md) — Stable Diffusion XL at 1024² native via sdcpp (default) or Comfy, with createImage + createVariant * [SD1 image generation](https://developer.civitai.com/orchestration/recipes/sd1.md) — classic Stable Diffusion 1.5 at 512² via sdcpp (default) or Comfy, with createImage + createVariant * [OpenAI image generation](https://developer.civitai.com/orchestration/recipes/openai.md) — GPT-Image 2 (default) / 1.5 / 1 and DALL·E 2 / 3 via OpenAI's hosted API * [Google image generation](https://developer.civitai.com/orchestration/recipes/google.md) — Imagen 4 and Nano Banana Pro / 2 / 2 Lite via Vertex AI, with editing + web-search grounding * [Krea v2 image generation](https://developer.civitai.com/orchestration/recipes/krea.md) — Krea v2 medium and large tiers via FAL, text-to-image * [Gemini image generation](https://developer.civitai.com/orchestration/recipes/gemini.md) — Gemini 2.5 Flash Image (same product as Nano Banana) via the direct Gemini API * [Seedream image generation](https://developer.civitai.com/orchestration/recipes/seedream.md) — ByteDance Seedream v3 / v4 / v4.5 / v5.0-lite with native up-to-4096 output + editing * [Grok image generation](https://developer.civitai.com/orchestration/recipes/grok.md) — xAI Grok with wide aspect-ratio menu (21 options) + editing * [WAN image generation](https://developer.civitai.com/orchestration/recipes/wan-image.md) — WAN v2.2 / v2.2-5b / v2.5 / v2.7 via FAL (image counterpart to the WAN video recipe) * [Image upscaling](https://developer.civitai.com/orchestration/recipes/image-upscaler.md) — ESRGAN-family upscalers, chain after `imageGen` or use standalone ## 3D * [3D model generation](https://developer.civitai.com/orchestration/recipes/3d.md) — text or image → 3D; Hunyuan3D on Civitai's Comfy workers (image-only, most control, cheapest — from 60 Buzz), Meshy v6 via FAL (text + image, rigging + animation, flat 1040 Buzz), and Tripo v2.5 via FAL (image-only, HD/PBR, quad topology, from 260 Buzz) ## Audio * [Transcription](https://developer.civitai.com/orchestration/recipes/transcription.md) — Qwen3-ASR, multilingual, word-level timestamps for captioning * [Text-to-speech](https://developer.civitai.com/orchestration/recipes/text-to-speech.md) — built-in speakers with optional style prompt, or voice cloning from a reference clip * [Multi-speaker dialogue](https://developer.civitai.com/orchestration/recipes/multi-speaker-dialogue.md) — overlay TTS clips on a shared timeline for debate, interview, or audio-drama scenes; the audio form of `composeMedia` * [ACE-Step music generation](https://developer.civitai.com/orchestration/recipes/ace-step-audio.md) — full songs from a style description + structured lyrics, 2B turbo default with optional 4B XL overrides; audio-only MP3 or MP4 with a still cover image ## Language models * [Chat completion](https://developer.civitai.com/orchestration/recipes/chat-completion.md) — any OpenRouter model or Civitai AIR model, vision inputs, tool use, streaming, image generation via `modalities: ["image"]`; OpenAI-compatible `/v1/chat/completions` endpoint or workflow step ## Utilities * [Prompt enhancement](https://developer.civitai.com/orchestration/recipes/prompt-enhancement.md) — LLM rewrites a user prompt for a target ecosystem (Flux / SDXL / SD1 / LTX2), returns issues + recommendations + enhanced prompt * [Image conversion](https://developer.civitai.com/orchestration/recipes/convert-image.md) — format conversion (JPEG / PNG / WebP / GIF), resize, and region blur; flat 1 Buzz * [Image background removal](https://developer.civitai.com/orchestration/recipes/image-background-removal.md) — BiRefNet subject cutout with a transparent background; PNG or WebP output, ~6 Buzz * [Web search](https://developer.civitai.com/orchestration/recipes/web-search.md) — SearXNG-backed web search with optional inline page scraping; results feed `repeat`/`$ref` composition or ground LLM answers * [Web scrape](https://developer.civitai.com/orchestration/recipes/web-scrape.md) — fetch and render a single page (headless browser) to LLM-ready markdown, HTML, or links ## Training Train a LoRA on your own dataset using AI Toolkit. You control training length with `steps`, the number of saved checkpoints with `epochs`, and can resume from an existing LoRA with `continueFrom`. All training runs are async — submit with `wait=0`. Cost is `steps × costPerStep + epochs × a per-epoch surcharge` with an 80%-of-default floor (rates vary per ecosystem). * [LoRA training overview](https://developer.civitai.com/orchestration/recipes/training.md) — what every ecosystem shares: parameters, training data (zip or uploaded blobs), continuing a run, live `trace` logs, the cost formula * [SDXL & SD1 LoRA training](https://developer.civitai.com/orchestration/recipes/training-sdxl-sd1.md) — classic Stable Diffusion ecosystems (500 Buzz default); cheapest pick for first fine-tunes * [Flux 1 LoRA training](https://developer.civitai.com/orchestration/recipes/training-flux1.md) — Flux.1 Dev or Schnell (2000 Buzz default); higher quality, fixed BFL base checkpoints * [Flux 2 Klein LoRA training](https://developer.civitai.com/orchestration/recipes/training-flux2-klein.md) — Flux 2 Klein 4b / 9b (500 / 1000 Buzz default), including image-edit training mode with control reference images * [Wan video LoRA training](https://developer.civitai.com/orchestration/recipes/training-wan.md) — preview ecosystem for Wan 2.1 / 2.2 video LoRAs (3000 Buzz default) * [LTX2 video LoRA training](https://developer.civitai.com/orchestration/recipes/training-ltx2.md) — Lightricks LTX2 and LTX 2.3 video LoRAs (2750 Buzz default, 3000-step runs) * [Chroma / ERNIE / Qwen / Z-Image LoRA training](https://developer.civitai.com/orchestration/recipes/training-other-image.md) — five smaller image ecosystems consolidated into one page; each section is independently runnable ::: tip Copy-paste runnable All recipes target `https://orchestration.civitai.com` and use `` as a placeholder for your Bearer token — an orchestration token ([Quick start → Prerequisites](https://developer.civitai.com/orchestration/guide/getting-started.md#prerequisites)). Drop them into curl, HTTPie, VS Code's REST Client, or any tool that speaks HTTP. ::: ## Common to every recipe Each recipe page documents only what is specific to it. This section covers what they all share. ### Reading results {#reading-results} A result arrives in the workflow envelope that [`GetWorkflow`](https://developer.civitai.com/orchestration/reference/operations/GetWorkflow.md) returns or a [webhook](https://developer.civitai.com/orchestration/guide/results-and-webhooks.md) delivers. Most generation recipes emit one of two output shapes. `imageGen` emits an `images[]` array, one entry per `quantity`. `nsfwLevel` carries the moderation classification applied to the output: ```json { "status": "succeeded", "steps": [{ "name": "0", "$type": "imageGen", "status": "succeeded", "output": { "images": [ { "id": "aa6e7228-68cd-4d15-b4d7-5005b2bfbac6-0.jpg", "width": 1024, "height": 1024, "url": "https://orchestration.civitai.com/v2/consumer/blobs/…?sig=…", "urlExpiresAt": "2027-04-15T17:18:54.3195353Z", "previewUrl": "https://orchestration.civitai.com/v2/consumer/blobs/…?sig=…", "previewUrlExpiresAt": "2027-04-15T17:18:54.3196735Z", "available": true, "nsfwLevel": "pg13" } ], "errors": [] } }] } ``` `videoGen` emits a single `video` blob: ```json { "status": "succeeded", "steps": [{ "name": "0", "$type": "videoGen", "status": "succeeded", "output": { "video": { "id": "blob_...", "url": "https://.../signed.mp4" } } }] } ``` Blob URLs (`url`, `previewUrl`) are signed and expire at `urlExpiresAt` — see [Blobs in results](https://developer.civitai.com/orchestration/guide/results-and-webhooks.md#blobs-in-results). ### Troubleshooting {#common-troubleshooting} These apply to every recipe; each page's own table covers the rest. For an expired blob URL, see [Errors & retries → Common gotchas](https://developer.civitai.com/orchestration/guide/errors-and-retries.md#common-gotchas). | Symptom | Likely cause | Fix | |---------|--------------|-----| | Request timed out (`wait` expired), or `202` with a non-terminal status | The job outlasted `wait` or [your HTTP client's timeout](https://developer.civitai.com/orchestration/guide/getting-started.md#_3-poll-if-you-didn-t-wait-inline) | Submit with `wait=0`, then [poll](https://developer.civitai.com/orchestration/guide/results-and-webhooks.md#polling-fallback) or use a [webhook](https://developer.civitai.com/orchestration/guide/results-and-webhooks.md). | | Step `failed`, `reason = "blocked"` | The prompt or an input hit content moderation (the provider's filter or Civitai's) | Don't retry the same input — see [Errors & retries → Step-level failures](https://developer.civitai.com/orchestration/guide/errors-and-retries.md#step-level-failures). | ### OpenAPI specs Each per-recipe endpoint publishes a standalone OpenAPI 3.1 YAML at `https://orchestration.civitai.com/v2/consumer/recipes//openapi.yaml` (for example [`imageGen`](https://orchestration.civitai.com/v2/consumer/recipes/imageGen/openapi.yaml)), ready to import into Postman, Insomnia or OpenAPI Generator. `` is the name in the recipe's `POST /v2/consumer/recipes/` endpoint (e.g. `videoGen`, `training`, `transcription`).