GLOW-UP aka lazy-WAN-KR (Florence Caption -> WAN2.2 T2I -> WAN2.2 I2V) — v0.8

Use GLOW-UP aka lazy-WAN-KR to turn an SD1.5 or SDXL image into one video clip: it captions the image with Florence, generates a new WAN2.2 image, and feeds that image into WAN2.2 image-to-video generation. You provide the source image, a WAN T2V target resolution, and optional .

Flow: Florence captioning → WAN2.2 text-to-image generation → WAN2.2 image-to-video generation.

Version v0.8 is described as a single-clip I2V back end; additional clips can be generated from the end frame with a multi-WAN-KR workflow.

The generated WAN2.2 image feeds the I2V stage, which produces a single video clip.

The WAN2.2 T2V stage additionally needs wan2.2_t2v_low_noise_14B_Q8_0.gguf; the description says WAN2.1 T2V should also work.

It is listed as a Workflows item using base model Wan Video 2.2 I2V-A14B, version v0.8.

Listed model files: Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64.safetensors; MiaoshouAI/Florence-2-large-PromptGen-v2.0; Wan2_1_VAE_fp32.safetensors; wan2.2_t2v_low_noise_14B_Q8_0.gguf; gimmvfi_r_arb_lpips_fp32.safetensors; Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1_high;

More listed model files: clip_vision_h.safetensors; Wan2.2-Lightning_I2V-A14B-4steps-lora_LOW_fp16.safetensors; OmniSR_X2_DIV2K.safetensors; umt5_xxl_fp16.safetensors; Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors; and Wan2.2-Lightning_I2V-A14B-4steps-lora_HIGH_

Listed node packs: comfyui-kjnodes, rgthree-comfy, comfyui-videohelpersuite, gguf, ComfyUI-GIMM-VFI, comfyui-aspect-ratio-crop-node, SeargeSDXL, comfyui-mxtoolkit, comfyui-custom-scripts, comfyui-florence2, comfyui-easy-use, and ComfyUI-GGUF.

Choose the WAN T2V target resolution as high as your machine permits, and add optional prompt directions to the Florence caption, such as camera movement or an action.

Suggestion · not verified

To inspect the T2I image before video generation, use the black group bypasser to block the video stage; if the image looks right, unblock GFXCARD_GOES_BRRR.

Suggestion · not verified

Before use, verify that these listed files are present: Wan2_1_VAE_fp32.safetensors, gimmvfi_r_arb_lpips_fp32.safetensors, Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1_high_noise_model.safetensors, clip_vision_h.safetensors, and OmniSR_X2_DIV2K.safetensors.

Suggestion · not verified

Also verify the exact node-pack names for ComfyUI-GIMM-VFI, comfyui-aspect-ratio-crop-node, comfyui-florence2, and ComfyUI-GGUF before use.

Suggestion · not verified