ComfyUI wan2.2: How to Create 1080p Videos with HuanYuan Video 1.5

You want clean 1080p AI video from still images. ComfyUI wan2.2 gives you flexible nodes and fast pipelines. Pair it with HuanYuan Video 1.5 for sharp motion, then add control tricks and smart upscaling. This guide shows setup, a working comfyui workflow, motion control, and low VRAM tips.

What is wan2.2 for ComfyUI?

It’s an AI video model setup inside ComfyUI that generates realistic short clips fast and plays well with control nodes and LoRA add‑ons.

• A modern AI video model build that runs inside ComfyUI
  
• Strong realism on faces and scenes
  
• Good speed with short clips
  
• Works well with control nodes and LoRA add-ons
  

Use it when you need quick drafts, product reels, or b-roll.

What do you need to get started?

Update ComfyUI, load Wan 2.2 and HuanYuan 1.5 models, and use a GPU near 12 GB VRAM with optional helper nodes like RGThree, Fusion LoRA, and Qwen.

• ComfyUI updated to the latest stable build
  
• A GPU with 12 GB VRAM or higher. Many users ask about “wan 2.2 12v.” Read it as 12 GB VRAM targets
  
• Model files for Wan 2.2 and HuanYuan Video 1.5 in your checkpoints folder
  
• Optional packs, Fusion LoRA, Qwen image edit 2509, RGThree quality-of-life nodes
  

Keep your install tidy. Name models clearly to avoid node errors.

How do you build a 1080p image-to-video workflow?

Load a clean image, run it through a video diffusion node with sane steps, then export frames to a 1080p MP4; keep sizes matched to the model.
Create a simple lane first, then add control.

1. Load Image. Use a clean 1024 or 1080 source
   
2. Encode latent. Match the model’s expected size
   
3. Video Diffusion node. Pick HuanYuan Video 1.5 or Wan 2.2
   
4. Sampler. 12 to 18 steps for drafts, 20 to 26 for finals
   
5. Frames to Video. Set 1080 × 1920 or 1920 × 1080
   
6. Save. Use H.264 or HEVC at high bitrate
   

Prompts

• Positive, subject, scene, lighting, lens, motion cue
  
• Negative, artifacts, extra limbs, blur, low res
  

How do you add motion and camera control?

Use camera language in the prompt, adjust motion strength, and add Fusion LoRA or San3 only lightly to avoid warping.
You shape the shot, then let the model fill the in-betweens.

• Use camera prompts. “slow dolly in,” “pan left,” “gentle orbit”
  
• Add motion strength sliders if your node set supports it
  
• Drop Fusion LoRA for style or steadier edges
  
• Try San3 for dynamic scenes that need punchier motion
  

Keep motion subtle. Overdriving leads to warping.

How do you upscale to clean 1080p?

Render smaller for speed and upscale with ESRGAN, or render native 1080p in short spans and stitch; sharpen once and raise bitrate.
Two reliable paths.

• Generate at 768 or 960, then upscale with a quality ESRGAN or 4x-UltraSharp node
  
• Or generate at 1080p in shorter lengths, then stitch sequences
  

Tips

• Sharpen only once at the end
  
• Raise bitrate. Target 20 to 40 Mbps for short clips
  
• Avoid double denoise passes
  

How do you get realistic image-to-video results fast?

Start with a strong photo, keep prompts steady, use short drafts before longer finals, and lock lighting so motion stays stable.

• Start from a strong photo. Fix faces first
  
• Use 12 to 16 frames for drafts, 24 to 48 for finals
  
• Keep prompts short and consistent across frames
  
• Lock lighting words. “soft studio,” “overcast,” “golden hour”
  
• Reduce seed changes. Stability beats variety here
  

How do you run low-VRAM builds cleanly?

Switch to half precision, cut context frames and batch size, disable extra previews, and render smaller then upscale.

• Use half precision where possible
  
• Lower context frames. 12 or 16 often looks fine
  
• Reduce batch size to 1
  
• Turn off preview nodes you do not need
  
• Split the job. Render halves and join later
  

If you still hit OOM, render at 768 first, upscale after.

How do RGThree labels help bigger workflows?

Labels and color‑coded groups make large graphs readable, reusable, and less error‑prone, so you wire once and reuse often.

• Add labels and colors to node groups
  
• Save reusable subgraphs for “Load → Diffuse → Save”
  
• Keep separate groups for “Prompt,” “Motion,” and “Output”
  
• You cut setup time and avoid wiring mistakes
  

How do you use multi-image reference for steadier style?

Feed face, body, and background images with balanced weights to lock character and tone while keeping motion flexible.

• Feed a face reference, a body shot, and a background
  
• Set lower weights on background to keep motion free
  
• Reuse the same set across scenes for a consistent character
  
• Lock color words to avoid tone drift
  

How do you refine frames with Qwen and Nano Banana?

Use Qwen to fix small frame issues and Nano Banana for light cleanup or style, keeping edits minimal to preserve motion.

• Qwen image edit 2509. Fix small issues frame by frame, like hands or text areas
  
• Nano Banana pro. Clean edges or stylize elements, then re-encode
  
• Keep changes light. Heavy edits per frame can break motion continuity
  

ComfyUI model quick compare

Mix and match. Draft in Wan 2.2, finalize in HuanYuan Video 1.5.

Sample comfyui workflow block

• Load Image → Encode → Video Diffusion (Wan 2.2) → Sampler 16 steps → Frames to Video 1080p
  
• Optional branch. Control camera prompt. Add Fusion LoRA
  
• Save MP4 at high bitrate
  

Save this as a preset. Clone it for each project.

Troubleshooting list

• Wobble on faces. Lower motion strength. Shorter clips help
  
• Soft detail. Upscale once. Raise bitrate. Add light sharpening
  
• VRAM errors. Lower resolution or steps. Use half precision
  
• Style drift. Reuse multi-image refs. Lock color and lighting words

Conclusion

ComfyUI wan2.2 gives you a fast lane to realistic clips. Build a simple comfyui workflow, add light camera control, and upscale to 1080p with care. Use HuanYuan Video 1.5 for final passes when you want extra detail. Keep prompts steady. Use multi-image reference for a consistent character. Label your graphs with RGThree to move faster. With the right setup, you get clean, controlled AI video that ships on time.

FAQs