(Video Tutorial Resources) Kontext, Speed Boosts and updated Loaders

This Flux.1 Kontext workflow is for image upscaling and image-to-video processing. It includes image and text-handling paths, with upscaled-image, preview, and combined-video outputs.

Its 180-node graph includes image loading, text encoding, scaling, sampling, VAE processing, upscaling, previews, and video combination.

The workflow uses Flux.1 Kontext and includes flux1-dev-kontext_fp8_scaled.safetensors from Flux1 Kontext Dev ComfyUI.

Image handling uses LoadImage; text paths use CLIPTextEncode and FramePackTimestampedTextEncode.

The graph includes ImageUpscaleWithModel for upscaling, PreviewImage for previews, and VHS_VideoCombine for video output.

Install FP8 or GGUF model files in models/checkpoints or unet; put Flux VAE in models/vae and T5 Clip FP16 in models/clip.

The notes also list NAG (ComfyManager #2706) and Hyper Chroma Speed Booster LORA.

The workflow lists these model files: 1x-ITF-SkinDiffDetail-Lite-v1.pth; bbox/face_yolov8m.pt; bj-w4-000007.safetensors; chroma-unlocked-v40.safetensors; clip_l.safetensors; flux1-dev-kontext_fp8_scaled.safetensors; flux1-kontext-dev_fp8_scaled_stochastic-marduk191.safetensors;

The list continues with FramePack_F1_I2V_HY_20250503_fp8_e4m3fn.safetensors; hunyuan_video_vae_bf16.safetensors; llava_llama3_fp16.safetensors; project0_real1smV3FP8.safetensors; sigclip_vision_patch14_384.safetensors; and t5xxl_fp16.safetensors.

Its 78 node types include Anything Everywhere?; BasicScheduler; Bookmark (rgthree); Bool; Cfg Literal; CLIPLoader; CLIPTextEncode; CLIPVisionEncode; CLIPVisionLoader; ConditioningZeroOut; CR Latent Input Switch; DetailerForEach; DiffusionModelLoaderKJ; DivideNode; DualCLIPLoader;

The accompanying notes point to Kontext Introduction, Kontext Setup, Kontext Speed Boosts, and Updated Flux and Chroma Loaders with Kontext.

Before running it, verify that EmptySD3LatentImage, CLIPVisionLoader, ReferenceLatent, and ConditioningZeroOut are available in ComfyUI.

Sugerencia · sin verificar

Also verify that t5xxl_fp16.safetensors, 1x-ITF-SkinDiffDetail-Lite-v1.pth, flux1-kontext-dev_fp8_scaled_stochastic-marduk191.safetensors, ImageScaleBy, and sigclip_vision_patch14_384.safetensors are present under those exact names in your ComfyUI setup.

Sugerencia · sin verificar