ComfyUI: Image-to-Video generation using Hunyuan 1.5 Distilled
ComfyUIDemo summary
The user demonstrates an I2V workflow in ComfyUI using the Hunyuan 1.5 distilled model, Google's Clip Vision for image encoding, and the 'image to video' node to animate a portrait.
Step-by-step
- Load the Hunyuan 1.5 Distilled main model at 720p resolution
- Set the Shift value to 7.0 and enable Sage Attention
- Resize the reference image to 720x1280 for portrait orientation
- Encode the reference image using the Google Clip Vision node
- Connect the VAE, start image, and Clip Vision output to the Hunyuan Video Image to Video node
- Link the generated positive/negative conditions and latent to the sampler
- Set the sampler to 20 steps and CFG to 1.0
- Upscale the resulting 720p video to 1080p using an enlargement model
Options
- Use Sage Attention to optimize performance
Watch out for
- The Hunyuan Video Image to Video node requires more connections than standard nodes, including VAE, start image, and CLIP vision output
- Do not reduce the sampling steps below 20 even when using the distilled model
Tips
- Use a CFG of 1.0 when working with the distilled version of the model
- Refer to the specific shift value list for optimal settings (e.g., 7.0 for this workflow)
Highlights
“Actually, this level of clarity is already quite high... Every detail is fantastic.”
All demos from “极简工作流！腾讯混元 1.5 (Hunyuan Video) 首发测评：原生 1080P + 显存优化全攻略”
- 5:282:26Text-to-Video generation with Hunyuan Video 1.5 in ComfyUIThe creator demonstrates a ComfyUI workflow using Hunyuan Video 1.5 (FP16) and Qwen2.5-VL for prompt generation to create a 720p video of a horse and rider.ComfyUI· AI Animation Generator
- 7:542:071080P Latent Upscaling with Hunyuan SR ModelThe video shows how to use the Hunyuan Video Latent Upscale node and the SR (Super Resolution) model to enhance a 720p video to 1080p using a dual-sampler noise injection technique.ComfyUI· AI Video Upscaler
- 10:012:01Image-to-Video generation using Hunyuan 1.5 DistilledCurrentThe user demonstrates an I2V workflow in ComfyUI using the Hunyuan 1.5 distilled model, Google's Clip Vision for image encoding, and the 'image to video' node to animate a portrait.ComfyUI· Image to Video
- Watch “极简工作流！腾讯混元 1.5 (Hunyuan Video) 首发测评：原生 1080P + 显存优化全攻略” →
Image to Video
7:500:28Image-to-Video action testThe creator demonstrates image-to-video capabilities by uploading a static image and prompting for an intense fight scene in both Hunyuan and Wan models.AI Search- 24:303:12Image-to-Video workflow in ComfyUIA demonstration of the image-to-video workflow in ComfyUI, including uploading a source image, setting crop dimensions, and generating an animated anime scene.AI Search
10:194:27Image-to-Video with LTX 2.3The creator demonstrates bringing a static image of a vampire warlord to life by using an image input node and a descriptive prompt to guide the animation and environmental effects.AIKnowledge2Go
6:160:38Image-to-video generation with LTX-2.3The creator demonstrates the image-to-video workflow in ComfyUI, showing how starting from an input image improves text legibility and subject consistency in the final video.MDMZ
5:244:57Image-to-video generation with Wan 2.2The user demonstrates how to use the Wan 2.2 GGUF model in ComfyUI to animate a static image of a woman based on a detailed text prompt.pixaroma
6:300:17Advanced multi-node AI workflowsThe video shows a complex Roboneo setup where an image is generated, passed through an image editing node, and finally converted into a video.Malva AI
3:070:38Image to Video with LTX DirectorThe creator demonstrates dragging an image into the LTX Director timeline and entering a text prompt to generate a video of a woman waving her hand.What Dreams Cost
6:130:36Animate posters using Luma and Kling nodesThe demonstration shows how to chain a video generation node (Luma and Kling/Seance) to the end of an image workflow to animate the final poster design.Sebastian Kamph
23:511:57Animate 3D renders with Wan 2.1The creator walks through a Wan 2.1 video generation workflow in ComfyUI, using the Painterly I2V node to control motion speed and animate a static 3D render.Matt Hallett Visual
0:171:00LTX-2 Image-to-Video distilled workflow setupThe demonstrator walks through loading the LTX-2 distilled checkpoint, upscaling models, and Gemma CLIP text encoder, while configuring resolution and frame rates for video generation.LTX
5:064:11Animate an image using Wan 2.1 in ComfyUIThe creator demonstrates loading a 'native image to video' workflow in ComfyUI, uploading a source image, and using the Wan 2.1 14B model to generate a high-quality video of a woman walking.Danish Sofi- 10:201:00Generate AI video of a horse with Wan 2.1A second demonstration showing the image-to-video process where a static image of a horse is animated into a walking sequence using the Wan 2.1 model in ComfyUI.Danish Sofi