WorkflowListedExternal

FLUX + CogVideoX + Hunyuan: T2I2V2V Simple Workflow

This is a simple multi model workflow I use to generate short videos.

TaskMultiCapabilitiesImage to video · Text to video · Video to videoFamilyHunyuanEstimated VRAMEstimate unavailableBgrade

Runtime profile

Access: External workflow
Status: Listed
Source: Civitai
Updated: Jul 20, 2026

Source description

This is a simple multi model workflow I use to generate short videos.

Text2Images2Videos2Video.

1 - FLUX to create 3 similar images from similar prompts

2 - CogVideoX 1.5 5B I2V to create two "interpolations" between those images

3 - Hunyuan V2V to solve inconsistencies and sudden changes in the video, specially in the background.

https://github.com/chengzeyi/Comfy-WaveSpeed

This version uses WaveSpeed to accelerate both FLUX and Hunyuan. Comfy probably won't find it as a Missing Node, make sure to follow install instructions from the link above.

AI-generated commentary

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

Purpose

Use this workflow to turn similar text prompts into short videos: FLUX creates three similar images, CogVideoX 1.5 5B I2V creates two interpolations between them, and Hunyuan V2V addresses inconsistencies and sudden background changes.

Flow

The described sequence goes from prompts to three images, then to two interpolations, followed by Hunyuan V2V processing.

Inputs and outputs

Input similar text prompts for FLUX.
Intermediate results are three similar images and two interpolations between them.
The final output is a short video processed with Hunyuan V2V.

Setup

Follow the Comfy-WaveSpeed installation instructions linked in the description; Comfy may not identify WaveSpeed as a missing node.
Suggestion · not verified

Dependencies

Listed model files: ae.safetensors, clip_l.safetensors, CogVideoX_5b_1_5_I2V_GGUF_Q4_0.safetensors, hunyuan_video_t2v_720p_bf16.safetensors, hunyuan_video_vae_bf16.safetensors, jibMixFlux_v72PixelHeaven.safetensors, llava_llama3_fp8_scaled.safetensors, and t5xxl_fp8_e4m3fn.safet‌
Node names include UNETLoader, VAEDecode, ModelSamplingSD3, VAEEncodeTiled, VAELoader, RandomNoise, GetNode, Seed (rgthree), Fast Groups Bypasser (rgthree), EmptySD3LatentImage, CLIPTextEncode, SetNode, ApplyFBCacheOnModel, ModelSamplingFlux, DownloadAndLoadCogVideoGGUFModel, and
Additional node names include FluxGuidance, ImageBatchMulti, VHS_SelectImages, DualCLIPLoader, KSamplerSelect, VAEDecodeTiled, CLIPLoader, CogVideoImageEncode, PurgeVRAMNode, CogVideoTextEncode, SamplerCustomAdvanced, CogVideoSampler, BasicGuider, VHS_VideoCombine, INTConstant, a

Configuration

The base model is Hunyuan Video, and the workflow combines FLUX, CogVideoX 1.5 5B I2V, and Hunyuan V2V stages.
This version uses WaveSpeed with both FLUX and Hunyuan.
This item is listed as a workflow, and no Pipe version is provided.

Cautions

Before use, check that the listed model files and node names are available.
Suggestion · not verified
Before use, account for WaveSpeed’s increased VRAM use; the description reports only a 24GB run.
Suggestion · not verified

Does this need to be edited?

Sources

1 source

CivitaiPrimary

civitai.com/models/1156702

Verified Jul 20

Source excerpts

1 excerpt

NoteJul 20civitai.com/models/1156702

Source context: 1012 downloads · Type Workflows · Base model Hunyuan Video