Download or deploy a Wan 2.2 5B photo-to-video workflow
Use the all-native ComfyUI graph locally, or deploy the same Photo Animator outcome on a managed cloud GPU without rebuilding the environment.
Start from a specific workflow outcome, review the expected GPU and cost profile, then deploy the hosted ComfyUI template when the fit is right.
No example clip is published for this download. Inspect the graph and its limitations before launching; its output quality and render time have not been demonstrated here on ModelPilot hardware.
- Workflow
- Wan 2.2 5B I2V
- Cloud GPU
- 48 GB tier
- First deploy
- 10–30 min
- Clip length
- 121 frames · ~5s
Download the ComfyUI workflow
Drag the JSON into ComfyUI. It uses native ComfyUI nodes only and defaults to 768×768, 121 frames, 24 fps and 30 steps. Its sampler ships with seed 869177064731501 set to randomize, so each run uses a new seed unless you set it to fixed.
Download workflow JSONImport this graph with your own image
The graph’s Load Image node points at sample_portrait.jpg, which is not included in the download, so ComfyUI reports a missing image until you select your own file or bypass that node. Unlike the upstream template, this file also carries no model-download metadata, so place the three files below manually. Using it on the hosted workspace is a manual step too: the deployment opens its own 1280×704, 81-frame graph, so drag this JSON onto that ComfyUI canvas to load these settings instead. No timed render on ModelPilot hardware is published for this exact graph.
Required model files
Use these exact filenames and directories when running locally.
| Role and file | Directory | Size |
|---|---|---|
Diffusion model wan2.2_ti2v_5B_fp16.safetensors | models/diffusion_models | ~6.4 GB |
Text encoder umt5_xxl_fp8_e4m3fn_scaled.safetensors | models/text_encoders | ~4.9 GB |
VAE wan2.2_vae.safetensors | models/vae | ~1.3 GB |
Direct answer
Can ModelPilot host a Wan 2.2 5B photo-to-video ComfyUI workflow?
Yes. ModelPilot hosts Photo Animator as an editable ComfyUI workspace and also publishes the all-native workflow JSON for local use.
Key facts
- iThe graph uses native ComfyUI nodes and no WanVideoWrapper.
- iThe Load Image node references sample_portrait.jpg, which is not shipped in the download; supply your own image or bypass that node.
- iThe published defaults are 768×768, 121 frames, 24 fps, and 30 steps; the hosted template opens at 1280×704 and 81 frames, so import this JSON there to match.
Next step
- 1Download the JSON when the user already has the three Wan model files.
- 2Use /deploy/photo-animator when the hosted template is the lower-friction path.
- 3Use /check first when the user has their own ComfyUI workflow JSON.
Not a fit when
- Keep motion prompts gentle; fast movement is where the 5B model deforms.
- Review model licenses separately before commercial use.
What this workflow includes
Inspect the graph, required files, and the current example limitations before choosing a hosted workspace.
The workflow is downloadable
The exact graph is available without signup, so local users can inspect and run it independently of ModelPilot.
Choose the configuration you want
The hosted workspace opens at 1280×704 for 81 frames. The download is 768×768 for 121 frames. Same nodes, same models, different canvas — import the JSON after launch if you want the downloaded configuration.
The limitations are explicit
The page states the practical five-second ceiling, softness tradeoff, and motion constraints before the deploy CTA.
End-to-end path
Move from a clear creative intent to a hosted workflow without rebuilding the ComfyUI environment from scratch.
- 1
Download or deploy
Use the JSON locally when the models are already installed, or start the hosted template.
- 2
Load a suitable photo
Images with hair, fabric, water, fog, or another subtle moving element produce more convincing motion.
- 3
Use a calm motion prompt
Describe gentle breathing, a soft breeze, or small facial motion rather than fast action.
- 4
Generate, review, then finish
Save the output, then manage the machine from your dashboard. Stopping ends GPU compute charges; delete the deployment when you no longer need its retained storage.
Use this workflow when
- You have portraits, product photos, or portfolio images to animate
- You want to inspect and edit the ComfyUI graph
- You want a short, gentle clip with stable facial and ambient motion
Choose another path when
- You need crisp, fast, large-amplitude movement; the 5B model tends to warp
- You need a single native clip much longer than about five seconds
- You require a custom model license or private-asset audit before launch
Frequently asked questions
Does this workflow require custom nodes?
No. The published Wan 2.2 5B graph uses native ComfyUI nodes and does not require WanVideoWrapper.
Can I use the workflow without ModelPilot?
Yes. Download the JSON and install the listed diffusion model, text encoder, and VAE in the corresponding ComfyUI model directories.
Why use the hosted version?
The hosted template is useful when you do not want to download roughly 12 GB of model files or manage a compatible GPU environment yourself.
Ready to deploy the workflow?
Deploy Photo Animator when you want the editable workflow without managing model downloads and GPU setup.