Unified CLI (opentryon)
Once OpenTryOn is installed (pip install -U opentryon or pip install -e .;
v0.0.9+ recommended), every adapter in the repo is available through a
single opentryon command with three levels of control: service →
model → parameters.
opentryon <service> --model <model> [params...]
- service: what kind of task --
vton,generate,edit,understand,video-generate,audio-to-video,generate-3d,bg-remove,tts,decide,embed,parse,classify,measure,detect,robot-act - model: which adapter/provider to use for that service, e.g.
--model flux-vto - parameters: model-specific flags (image inputs, prompts, sampling knobs, etc.)
Services and Models
| Service | What it does | Models |
|---|---|---|
vton | Virtual try-on: compose a garment onto a person image | flux-vto, google-vton, outfitanyone-plus, photoroom-vton, photoroom-virtual-model, nova-canvas, kling-ai, segmind, p-image-tryon, fashn-tryon-max, fashn-tryon-v1.6, nano-banana-2-lite, qwen-image, qwen-image-local, leffa (local), catvton (local), fashn-vton-1.5 (local), muse-image, klein-4b-tryon (local) |
generate | Text-to-image generation | nano-banana, nano-banana-pro, nano-banana-2, nano-banana-2.1, flux3-image, flux3-image-klein, flux3-image-klein-local (local), flux2-pro, flux2-flex, flux2-turbo (local), sana-sprint (local), gpt-image, gpt-image-2.5, gpt-image-2.5-sunburst, luma-image, seedream, ideogram, recraft-v3, grok-imagine-image, p-image, p-image-ideogram, qwen-image, qwen-image-2.1-pro, qwen-image-2.1-turbo, qwen-image-local (local), qwen-image-2.1-local, qwen-image-2.1-turbo-local (local), muse-image, fashn-model-create |
edit | Image editing (image + instruction → image) | nano-banana, nano-banana-pro, nano-banana-2, nano-banana-2.1, flux3-image, flux3-image-klein, flux3-image-klein-local (local), flux2-pro, flux2-flex, flux2-turbo (local), gpt-image, gpt-image-2.5, gpt-image-2.5-sunburst, seedream, recraft-v3, p-image-edit, p-image-upscale, qwen-image, qwen-image-2.1-pro, qwen-image-2.1-turbo, qwen-image-local (local), qwen-image-2.1-local, qwen-image-2.1-turbo-local (local), muse-image, fashn-product-to-model, fashn-model-swap, fashn-face-to-model, fashn-reframe, fashn-edit, qie-2511-extract-outfit (local) |
understand | Image/video understanding | kimi-k2.6, kimi-k2.7-code, kimi-k3, kimi-k3.1-pro, ripple-65b-openrouter, ripple-65b-local, kimi-vl (local), qwen3.8-max, qwen3.8-omni-flash, qwen3.8 (local), glm-5.3-flashx, deepseek-flash, deepseek-vl2 (local), deepseek-ocr (local), claude-haiku-5-5, gemini-3.8-flash, hy4-preview, hy4-preview-local, nemotron-omni, nemotron-3.5-super-vl, cosmos3-reasoner, llava-next (local), ternary-bonsai-2-27b (local server), limite-1b-violetto (local, text-only math reasoning — out of fashion/media scope), beam (Reflection, text-only), laguna-xs.2 / laguna-xs.2.1 / laguna-m.1 (Poolside, text-only), trinity-large / trinity-large-thinking / trinity-mini (Arcee, text-only), trinity-nano (local server), flint-local, flint-openrouter, sage-a1.5-local, fashion-florence (local) |
video-generate | Text/image-to-video generation | veo, sora, luma-video, luma-ray-3.2, flux3-video, flux3-max, seedance, kling-v4, kling-v3, kling-v3-omni, kling-v2-5-turbo, grok-imagine-video, gemini-omni, p-video, p-video-2-pro, p-video-replace, p-video-avatar, p-video-animate, ltx-2.5-api, ltx-2.5, hailuo-2.3, minimax-h3, minimax-h3-max, fal-h3-max, fal-h3-max-lipsync, fal-h3-max-relight, minimax-h3-local, wan-api, wan-3.0, wan-2.2, runway-gen4.5, cosmos3, cosmos3-local (local), fashn-image-to-video |
embed | Multimodal embeddings | … includes nemotron-3.5-embed-1b, nemotron-3.5-embed-1b-local, gemini-embedding-2, … |
parse | Document/page image parsing | nemotron-parse, nemotron-parse-local, lightonocr-3-local (local), lightonocr-3-vllm |
audio-to-video | Lip-sync / dub video to speech audio | nvidia-lipsync (self-hosted NIM gRPC) |
generate-3d | Text/image to 3D mesh | trellis (NVIDIA NIM) |
bg-remove | Background removal | ben2 (local), fashn-bg-remove, recraft-v3 |
tts | Text-to-speech | eleven-v4, eleven-v4-turbo, gemini-3.8-flash-tts, gemini-3.8-flash-lite-tts |
decide | Calibrated yes/no, choice, score, box, and track decisions ("System One" models, no text output) | jev, pplx-decider, pplx-decider-local, clef, clef-local, kev-openrouter, kev-local, autotrust-jev-local, thx-01, thx-01-local (local), d1-3b (local), d1-omni-600m (local), d3-turbo … d3-edge, d3-*-local (local), videojev-4b (local), videojev-35b-a3b (local) |
classify | Fixed-label image classification with calibrated probabilities (try-on safety gating) | tryon-safety-classifier (local) |
measure | Body measurements (ISO 8559-1 sizing) | clad-body-photo, clad-body-questionnaire (Clad API), clad-body-local (local Anny/MHR params) |
detect | Open-vocabulary boxes, points, or an SVG mask | moondream-3.1, moondream-3-preview, moondream-3.1-local |
robot-act | Robot policy actions (World Action Models) | dream-1, dream-1-flash (DREAM_API_KEY, install dreamlabs-client) |
embed | Multimodal embeddings saved to .npz | gemini-embedding-2, embeddinggemma-2 (local), pplx-embed-v2-late-0.6b (local), pplx-embed-v2-late-9b (local), marqo-fashionsiglip (local), moda-fashion-distilled (local), moda-pro-lite (local), zooclaw-fashionsiglip2 (local) |
Models marked "local" run on your own GPU and require
pip install opentryon[local]; everything else calls a cloud API and needs
the corresponding API key set in your environment (see
Configuration).
Discovering Flags
Every level of the CLI is self-documenting:
opentryon --help # list services
opentryon understand --help # list models for a service
opentryon understand --model kimi-k2.6 --help # list that model's parameters
Examples
# Virtual try-on
opentryon vton --model flux-vto \
--person-image model.png --garment-image garment.png
# Google Vertex dedicated try-on (ADC + GOOGLE_CLOUD_PROJECT; not GEMINI_API_KEY)
opentryon vton --model google-vton \
--person-image model.png --garment-image garment.png
# Alibaba OutfitAnyone-Plus (Beijing DASHSCOPE_API_KEY; not Qwen-Image)
opentryon vton --model outfitanyone-plus \
--person-image model.png --garment-image garment.png
# Photoroom shopper try-on / catalog virtual model
opentryon vton --model photoroom-vton \
--person-image selfie.jpg --garment-image dress.jpg
opentryon vton --model photoroom-virtual-model \
--garment-image flatlay.jpg --preset-model avery
# Text-to-image
opentryon generate --model nano-banana-pro \
--prompt "A fashion model wearing elegant evening wear" --resolution 4K
# Image editing
opentryon edit --model gpt-image-2.5-sunburst \
--images person.jpg --prompt "Change the jacket to black leather"
# Image/video understanding (Kimi K2.6, general-purpose -- not fashion-only)
opentryon understand --model kimi-k2.6 \
--image garment.jpg --prompt "Describe this outfit."
opentryon understand --model kimi-k2.6 \
--video runway_clip.mp4 --prompt "Summarize the styling shown."
# Coding-focused multimodal understanding
opentryon understand --model kimi-k2.7-code \
--image ui_mockup.png --prompt "Write the HTML/CSS for this design."
# Kimi K3 (flagship multimodal reasoning)
opentryon understand --model kimi-k3 \
--image garment.jpg \
--prompt "Write a marketplace-ready title and bullet points." \
--reasoning-effort high
# Kimi K3.1 Pro (post-trained K3 successor; --kimi-model kimi-k3.1-pro-highspeed for the faster variant)
opentryon understand --model kimi-k3.1-pro --image garment.jpg --prompt "Describe this outfit." --reasoning-effort high
# Open-weight local understanding (no API key, needs a GPU)
opentryon understand --model kimi-vl --image garment.jpg
# Qwen3.8-Max (DashScope) + open-weight Qwen3.8-27B
# Native multimodal understand: thinking + reasoning_effort on Max
opentryon understand --model qwen3.8-max \
--image garment.jpg --prompt "Describe this outfit." \
--reasoning-effort medium
opentryon understand --model qwen3.8 --image garment.jpg
# Qwen3.8-Omni-Flash (same DASHSCOPE_API_KEY): adds --audio
opentryon understand --model qwen3.8-omni-flash \
--audio voice_note.wav --prompt "What is being said?"
# GLM-5.3-FlashX (Zhipu / Z.ai, ZAI_API_KEY)
opentryon understand --model glm-5.3-flashx \
--image garment.jpg --prompt "Describe this outfit."
# DeepSeek-V4.1-Flash (DEEPSEEK_API_KEY; image only, no video)
opentryon understand --model deepseek-flash \
--image garment.jpg --prompt "Describe this outfit." --reasoning-effort max
# DeepSeek-VL2 / DeepSeek-OCR (local, no first-party hosted API exists for either)
opentryon understand --model deepseek-vl2 --image garment.jpg --prompt "Describe this outfit."
opentryon understand --model deepseek-ocr --image care_label.jpg --ocr-mode markdown
# Ternary Bonsai 2 27B (local llama.cpp/MLX server, no cloud key)
opentryon understand --model ternary-bonsai-2-27b \
--prompt "Explain ternary quantization in two sentences."
# Limite 1B - Violetto (local, text-only math reasoning -- out of fashion/media scope)
opentryon understand --model limite-1b-violetto \
--prompt "If x + 3 = 8, what is x?"
# Reflection Beam (REFLECTION_API_KEY; text only, no image)
opentryon understand --model beam \
--prompt "Write a 3-sentence lookbook caption for a linen trench."
# Poolside Laguna (POOLSIDE_API_KEY; text only)
opentryon understand --model laguna-m.1 \
--prompt "Write a 3-sentence lookbook caption for a linen trench."
# Arcee Trinity Large 400B (ARCEE_API_KEY; text only)
opentryon understand --model trinity-large \
--prompt "Write a 3-sentence lookbook caption for a linen trench."
# Tencent Hy4 preview (TokenHub LLM; optional --image)
opentryon understand --model hy4-preview \
--prompt "Write a 3-sentence lookbook caption for a linen trench."
# Local vLLM/SGLang twin (HY4_BASE_URL, no TokenHub key)
opentryon understand --model hy4-preview-local --prompt "Hello"
# NVIDIA NIM (same NVIDIA_API_KEY): Nemotron Omni + Cosmos 3 Reasoner
opentryon understand --model nemotron-omni \
--image garment.jpg --prompt "Describe this outfit."
opentryon understand --model cosmos3-reasoner \
--video lookbook.mp4 --prompt "What physical interactions occur?"
# Qwen-Image 3.0 (same DASHSCOPE_API_KEY): generate / edit / VTON
opentryon generate --model qwen-image \
--prompt "editorial lookbook, linen trench on a sunlit terrace"
opentryon edit --model qwen-image \
--images person.jpg --prompt "Change the jacket to black leather"
opentryon vton --model qwen-image \
--person-image model.jpg --garment-image garment.jpg \
--garment-description "olive green bomber jacket"
# Qwen-Image local Diffusers (needs GPU + opentryon[local])
opentryon generate --model qwen-image-local \
--prompt "editorial lookbook, linen trench" --aspect-ratio 16:9
opentryon vton --model qwen-image-local \
--person-image model.jpg --garment-image garment.jpg
# Dedicated local VTON (needs GPU + opentryon[local])
opentryon vton --model leffa --person-image model.jpg --garment-image garment.jpg
opentryon vton --model catvton --person-image model.jpg --garment-image garment.jpg
# Apache-2.0 maskless local VTON (pip install 'git+https://github.com/fashn-AI/fashn-vton-1.5.git')
opentryon vton --model fashn-vton-1.5 --person-image model.jpg --garment-image garment.jpg --category tops
# Nano Banana 2.1 (GEMINI_API_KEY): successor to nano-banana-2, adds 1:4 / 4:1 / 1:8 / 8:1
opentryon generate --model nano-banana-2.1 --prompt "Panoramic lookbook banner" --aspect-ratio 4:1
# Text-to-video
opentryon video-generate --model veo \
--prompt "A model walking a runway in slow motion" --duration 6
# Veo 3.1: first+last frame, reference images, 4k (8s), Lite tier
opentryon video-generate --model veo --prompt "Transition" --image a.jpg --last-image b.jpg --duration 8
opentryon video-generate --model veo --prompt "Walk" --reference-image a.jpg b.jpg --duration 8
opentryon video-generate --model veo --prompt "Walk" --duration 8 --resolution 4k
# Gemini Omni 1.2 Flash: extend an existing clip (or --duration 3-15 for a longer new clip)
opentryon video-generate --model gemini-omni --prompt "Continue the walk" --video clip.mp4
opentryon video-generate --model flux3-video --prompt "Runway walk, soft light" --duration 8 --resolution fhd
opentryon video-generate --model flux3-video --prompt "Turn toward camera" --image look.jpg
# Seedance 2.5 reference-to-video
opentryon video-generate --model seedance --prompt "Repeat the motion" --reference-image p.jpg --reference-video m.mp4 --duration 10
# Local Cosmos 3 (GPU)
opentryon video-generate --model cosmos3-local --prompt "Runway at dusk"
# NOTE: sora is deprecated -- OpenAI removed the API on 2026-09-24
# Seedance 2.5 / Kling 3.0 / Ray 3.2 / Grok Imagine Video / Pruna P-Video
opentryon video-generate --model seedance --prompt "10s lookbook walk" --duration 10
# Seedance 2.5 ModelArk id is dreamina-seedance-2-5-260628 (--model-version seedance-2-5)
opentryon video-generate --model kling-v3 --prompt "atelier pan" --mode pro --sound on
opentryon video-generate --model luma-ray-3.2 --prompt "dolly through mist" --resolution 720p
opentryon video-generate --model grok-imagine-video --prompt "cinematic push-in" --duration 6
opentryon video-generate --model p-video --prompt "runway walk, soft light" --duration 5
opentryon video-generate --model p-video-2-pro --prompt "runway walk, soft light" --duration 8 --mode speed
opentryon video-generate --model p-video-replace --video clip.mp4 --images identity.jpg
opentryon video-generate --model p-video-avatar --image portrait.jpg --voice-script "Hello from the showroom."
opentryon video-generate --model p-video-animate --video driver.mp4 --image subject.jpg
opentryon video-generate --model ltx-2.5-api --prompt "runway walk, soft light" --duration 8 --resolution 1920x1080
opentryon video-generate --model ltx-2.5 --prompt "runway walk at dusk, camera tracking" --width 960 --height 544 --num-frames 121
opentryon video-generate --model hailuo-2.3 --prompt "runway walk [Tracking shot]" --duration 6 --resolution 1080P
opentryon video-generate --model minimax-h3 --prompt "runway walk at dusk" --duration 5 --resolution 2K --ratio 16:9
opentryon video-generate --model minimax-h3-max --prompt "runway walk at dusk" --duration 5 --resolution 768P --ratio 16:9
opentryon video-generate --model fal-h3-max --prompt "runway walk at dusk" --duration 5 --resolution 768P --ratio 16:9
opentryon video-generate --model fal-h3-max --prompt "Image 1 is the model. Keep her identity." --reference-image look.jpg
opentryon video-generate --model fal-h3-max-lipsync --image portrait.jpg --audio line.wav --resolution 1080P
opentryon video-generate --model minimax-h3-local --prompt "runway walk at dusk" --width 960 --height 544 --num-frames 124
opentryon video-generate --model wan-api --prompt "runway walk" --duration 5 --resolution 720P
opentryon video-generate --model wan-3.0 --prompt "runway walk at dusk" --duration 8 --resolution 720P
opentryon video-generate --model wan-2.2 --prompt "runway walk at dusk" --num-frames 81
opentryon video-generate --model runway-gen4.5 --prompt "runway walk through mist" --duration 5 --ratio 1280:720
opentryon video-generate --model cosmos3 --prompt "A model walks a concrete runway at dusk." --resolution 720
# Seedream / Ideogram / Grok Imagine Image / Pruna P-Image
opentryon generate --model seedream --prompt "editorial sneaker still" --size 2K
opentryon generate --model ideogram --prompt 'Poster "SUMMER 2026"' --rendering-speed QUALITY
opentryon generate --model recraft-v3 --prompt "linen trench on a terrace" --style Photorealism --size 3:4
opentryon edit --model recraft-v3 --image look.jpg --prompt "the same scene in winter" --strength 0.4
opentryon bg-remove --model recraft-v3 --image product.jpg
opentryon generate --model grok-imagine-image --prompt "street-art collage" --aspect-ratio 16:9
opentryon generate --model p-image --prompt "luxury knitwear flatlay" --aspect-ratio 1:1
opentryon generate --model p-image-ideogram --prompt 'Poster "ATELIER NOIR"' --thinking high --image-size 2K
opentryon generate --model muse-image --prompt "editorial runway still, dusk" --size 1024x1536
opentryon edit --model muse-image --images look.jpg --prompt "black leather jacket, keep pose"
opentryon vton --model muse-image --person-image model.jpg --garment-image garment.png
opentryon edit --model p-image-edit --images photo.jpg --prompt "clean studio background"
opentryon edit --model p-image-upscale --image photo.jpg --target 8
# Background removal
opentryon bg-remove --model ben2 --image product.jpg --refine
# Claude Haiku 5.5 understanding (ANTHROPIC_API_KEY)
opentryon understand --model claude-haiku-5-5 --image garment.jpg --prompt "Fabric, colour, fit?" --effort low
# H3 Max Relight (FAL_KEY): re-light a clip from a lighting-sphere render
opentryon video-generate --model fal-h3-max-relight --video runway.mp4 --reference-image sphere.png
# Decisions: typed, calibrated answers (questions.json: see the Jev docs page for the schema)
opentryon parse --model lightonocr-3-local --image size_chart.pdf --pages 1-2 --grounding # local OCR -> markdown + boxes
opentryon decide --model jev --questions questions.json --state "Red satin evening gown" # TYPESAFE_API_KEY
opentryon decide --model d1-3b --questions questions.json --image dress.jpg # local GPU
opentryon decide --model pplx-decider --questions questions.json --image dress.jpg # PERPLEXITY_API_KEY
opentryon decide --model clef --clef-model clef-flash --questions questions.json --state "..." # CLOUDFLARE_API_TOKEN + CLOUDFLARE_ACCOUNT_ID
opentryon decide --model kev-openrouter --questions questions.json --state "..." # OPENROUTER_API_KEY
opentryon decide --model videojev-4b --video clip.mp4 --questions videojev_tasks.json # local GPU (VideoJev schema)
# FASHN catalogue endpoints (same FASHN_API_KEY as fashn-tryon-max)
opentryon edit --model fashn-product-to-model --image jacket.jpg --prompt "studio background" --resolution 2k
opentryon edit --model fashn-model-swap --model-image look.jpg --prompt "woman with red hair"
opentryon edit --model fashn-reframe --image look.jpg --aspect-ratio 16:9
opentryon video-generate --model fashn-image-to-video --image look.jpg --duration 5 --resolution 720p
# Gate a product before try-on (pass ALL its photos; local)
opentryon classify --model tryon-safety-classifier --image front.jpg back.jpg
# Body measurements for sizing (Clad API — CLAD_API_KEY)
opentryon measure --model clad-body-photo --photo full_body.jpg --height-cm 172
# Ask Moondream about a look, or box the garment (MOONDREAM_API_KEY)
opentryon understand --model moondream-3.1 --image look.jpg --prompt "What fabric is the jacket?"
opentryon detect --model moondream-3.1 --image look.jpg --object dress
# Try-on on the Apache-2.0 FLUX.2 klein 4B LoRA (person must be clothing-agnostic, or pass --mask-image)
opentryon vton --model klein-4b-tryon --person-image person_agnostic.png --top-image shirt.png --bottom-image jeans.png
# Garment tagging and extraction (local GPU)
opentryon understand --model fashion-florence --image dress.jpg
opentryon edit --model qie-2511-extract-outfit --image model_wearing_look.jpg
# Fashion catalogue search (local): by text or by photo
opentryon embed --model marqo-fashionsiglip --image catalogue/*.jpg --query "black leather ankle boots"
# Embeddings (hosted, GEMINI_API_KEY): text/image/audio/video/PDF in one space; --aggregate fuses inputs
opentryon embed --model gemini-embedding-2 --image boot1.jpg boot2.jpg --query "black leather ankle boots" --dim 768
# Embeddings (local GPU): vectors -> outputs/*.npz, --query adds similarity scores
opentryon embed --model embeddinggemma-2 --text "red gown" --image dress.jpg --query "formal dress"
opentryon embed --model pplx-embed-v2-late-0.6b --image page1.png page2.png --query "wool coat"
# Gemini 3.8 Flash understanding: images, video, audio and PDFs in one call (GEMINI_API_KEY)
opentryon understand --model gemini-3.8-flash --video runway.mp4 --pdf size_chart.pdf --prompt "Summarise the looks and sizes" --thinking-level low
# Gemini TTS: 30 voices, --style for delivery, two-speaker dialogue
opentryon tts --model gemini-3.8-flash-tts --text "Welcome to the spring collection." --voice Puck --style "warm and upbeat"
opentryon tts --model gemini-3.8-flash-lite-tts --dialogue turns.json --speaker Joe=Puck Jane=Kore
# Text-to-speech (ELEVENLABS_API_KEY)
opentryon tts --model eleven-v4 --text "Welcome to the spring collection."
opentryon tts --model eleven-v4-turbo --text "[whispers] New drop just landed." \
--voice-id 21m00Tcm4TlvDq8ikWAM --output-format wav_44100
Every command accepts -o/--output-dir (default: outputs/) and
--dry-run (print the resolved adapter call without invoking the API/GPU):
opentryon vton --model flux-vto \
--person-image model.png --garment-image garment.png --dry-run
Local (GPU-only) Models
Local models (fashn-vton-1.5, flux2-turbo, kimi-vl, qwen3.8, deepseek-vl2, deepseek-ocr, limite-1b-violetto, qwen-image-local, leffa, catvton, llava-next, ben2, ltx-2.5, minimax-h3-local, wan-2.2) need the
local extra:
pip install opentryon[local]
Running a local model without it prints an install hint instead of a raw stack trace:
✗ 'Kimi-VL (open-weight, local)' requires local ML dependencies that aren't installed.
Install them with: pip install opentryon[local]
See Also
- MCP Server
- TryOn Studio
- OpenAPI & Postman
- Kimi K2.6 / K2.7 Code / K3 understanding
- Kimi-VL open-weight local model
- Qwen3.8-Max / Qwen3.8-Omni-Flash understanding
- GLM-5.3-FlashX understanding
- DeepSeek-V4.1-Flash understanding
- DeepSeek-VL2 (local, open-weight)
- DeepSeek-OCR (local, open-weight)
- Ternary Bonsai 2 27B (local server)
- Limite 1B - Violetto (local, out of fashion/media scope)
- ElevenLabs Eleven v4 / v4 Turbo (text-to-speech)
- Gemini 3.8 Flash (understanding + TTS) · Claude Haiku 5.5 · Typesafe Jev · Decider, Clef, Kev, JEV
- Gemini Embedding 2 · Liquid d1 · EmbeddingGemma 2 · pplx-embed-v2-late
- Hy4 preview TokenHub
- Hy4 local vLLM/SGLang
- Qwen-Image generation, edit, and try-on
- OutfitAnyone-Plus dedicated try-on
- Photoroom Virtual Try-On / Virtual Model
- Qwen-Image open-weight local model
- Qwen3.8 open-weight local model
- Adding a new model to the CLI
- Roadmap