WorkflowListedExternal

Wan2.1 Upscaler + FrameGen — v2.0

Added image upscale using model instead of lanzcos.

VersionRelease 2098578 — wan2.1_upscale+frame-gen-v2.0.jsonEstimated VRAMNot yet characterizedUnratedgrade

Runtime profile

Source description

Added image upscale using model instead of lanzcos.

Special thx to AI Artistry Atelier. I borrowed an idea from his video on youtube.

Remaining hardware or download-size estimates are not available yet.

Requirements

KijaiWan21T2V13B.safetensors
Not resolved
Diffusion model · Hugging Face · Kijai/WanVideo_comfy
RealESRGAN_x2plus.pth
Found
Upscaler · 64.0 MB · PTH · Unknown
rife47.pth
Repository matched
Checkpoint · Unknown
umt5_xxl_fp8_e4m3fn_scaled.safetensors
Found
Text encoder · 6.27 GB · SAFETENSORS · U8 + FP16 + FP32 + FP8 (tensor header) · Hugging Face · chatpig/encoder
wan_2.1_vae.safetensors
Found
VAE · 242 MB · SAFETENSORS · BF16 (tensor header) · Hugging Face · Comfy-Org/Wan_2.1_ComfyUI_repackaged
Wan21_CausVid_bidirect2_T2V_1_3B_lora_rank32.safetensors
Not resolved
LoRA · Hugging Face · Kijai/WanVideo_comfy

AI-generated commentary

AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.

Use this workflow to upscale video and generate frames. It loads a video, applies model-based frame upscaling and RIFE VFI processing, then combines the frames into an output video.

The video path includes VHS_LoadVideo and VHS_VideoCombine, alongside ImageResizeKJv2, ImageUpscaleWithModel, and RIFE VFI nodes.

The graph also includes CLIP text encoding, Wan model loading, and sampling nodes.

The package lists these model files: KijaiWan21T2V13B.safetensors, RealESRGAN_x2plus.pth, rife47.pth, umt5_xxl_fp8_e4m3fn_scaled.safetensors, wan_2.1_vae.safetensors, and Wan21_CausVid_bidirect2_T2V_1_3B_lora_rank32.safetensors.
It also lists these required ComfyUI node packs: comfyui_essentials, comfyui-frame-interpolation, comfyui-kjnodes, and comfyui-videohelpersuite.

For v2.0, download an upscaler model of your choice; the release description says it adds model-based image upscaling.
Suggestion · not verified

The author reports about 160–480 seconds to scale and generate frames for a 33-frame video after models load, depending on resolution, and says they used an RX 6800 XT.

Does this need to be edited?

Sources

1 source

CivitaiPrimaryVersion 2098578

Verified Nov 3