Uploaded April 2026 | Updated November 2026, 4 days ago
My new ERNIE Image ComfyUI workflow generates perfect text and complex diagrams every single time. I will show you exactly how to run this fast, even if your computer relies on a low-VRAM graphics card.
We need to talk about the Prompt Enhancer tool. You must use this feature when you generate flow diagrams, movie posters, and comic pages. You must turn it off when you want realistic photos. If you leave it on, you get subjects with artificial, plastic skin. Disabling the enhancer forces the model to create authentic, natural human faces.
I tested ERNIE Base, ERNIE Turbo, and Z-Image-Turbo. ERNIE Turbo won the test, but you have to format your text properly to get these exact results. Do not write a normal paragraph. You need to structure your prompt exactly like a strict design brief. This specific format stops the AI from repeating random words.
You should use ERNIE Turbo for your most complex generations. We use it to build full comic pages, character sheets, and diagrams with directional arrows. It also works perfectly for movie posters and gaming UI.
This model truly understands visual storytelling. I gave it a prompt for a Viking storyboard, and it generated three distinct moments featuring the exact same character with naturally changing emotions.
If you have low video memory, simply use the GGUF version. It runs completely smooth on budget hardware. I also added RTX Video Super Resolution into our chain. This tool takes your base image and sharpens the tiny details so your small text becomes incredibly easy to read.
Workflow: aistudynow.com/ernie-comfyui-workflow-use-to-generate-perfect-text
X Profile: https://x.com/aistudynowcom
My new ERNIE Image ComfyUI workflow generates perfect text and complex diagrams every single time. I will show you exactly how to run this fast, even if your computer relies on a low-VRAM graphics card.
We need to talk about the Prompt Enhancer tool. You must use this feature when you generate flow diagrams, movie posters, and comic pages. You must turn it off when you want realistic photos. If you leave it on, you get subjects with artificial, plastic skin. Disabling the enhancer forces the model to create authentic, natural human faces.
I tested ERNIE Base, ERNIE Turbo, and Z-Image-Turbo. ERNIE Turbo won the test, but you have to format your text properly to get these exact results. Do not write a normal paragraph. You need to structure your prompt exactly like a strict design brief. This specific format stops the AI from repeating random words.
You should use ERNIE Turbo for your most complex generations. We use it to build full comic pages, character sheets, and diagrams with directional arrows. It also works perfectly for movie posters and gaming UI.
This model truly understands visual storytelling. I gave it a prompt for a Viking storyboard, and it generated three distinct moments featuring the exact same character with naturally changing emotions.
If you have low video memory, simply use the GGUF version. It runs completely smooth on budget hardware. I also added RTX Video Super Resolution into our chain. This tool takes your base image and sharpens the tiny details so your small text becomes incredibly easy to read.
Workflow: aistudynow.com/ernie-comfyui-workflow-use-to-generate-perfect-text
X Profile: https://x.com/aistudynowcom








![WAN 2.2 Animate in ComfyUI: InfiniteTalk + Uni3C — Low VRAM (GGUF Q4) [Free Workflow]
Download the free workflow: https://aistudynow.com/wan-2-2-animate-comfyui-infinitytalk-uni3c-q4-low-vram/
#ComfyUI #WAN22 #InfinityTalk
Swap any face or object inside one ComfyUI flow. Blend looks real. Lips match. Camera move support with Uni3C. FP8 vs GGUF Q4 test on low VRAM.
what you will learn
InfiniteTalk with WAN 2.2 Animate for better lip sync
Uni3C camera motion test and when to keep it off
FP8 vs Q4 GGUF on low VRAM and what to pick
LoRA we used
Relighting LoRA for Animate to match scene color and tone WanAnimate_relight_lora_fp16.safetensors
LightX 4-step LoRA set for Animate or I2V
low_noise_model.safetensors
Pusa v1 motion LoRA if you want stronger motion : Wan21_PusaV1_LoRA_14B_rank512_bf16.safetensors
HPS LoRA for human-preference quality tuning : Wan2.2-Fun-A14B-InP-LOW-HPS2.1_resized_dynamic_avg_rank_15_bf16.safetensors
upload the edit video and pick a supported WAN 2.2 size with Resolution Master
upload your subject image at the same size, remove background
mask with points: green keep, red remove, test mask once
models:
Wan2.2-Animate-14B FP8 model: Wan2_2-Animate-14B_fp8_e4m3fn_scaled_KJ.safetensors
Folder: ComfyUI/models/diffusion_models
Optional for low VRAM: Wan2.2 I2V GGUF Q4 or Q5 pairs here is the link
disclosure and credits
this video uses AI to alter real footage for an educational demo. altered or synthetic content label is on in Studio.
WAN, InfinityTalk, Uni3C, LoRA models belong to their creators. credits to them. WAN 2.2 Animate in ComfyUI: InfiniteTalk + Uni3C — Low VRAM (GGUF Q4) [Free Workflow]](https://i.ytimg.com/vi/fB7rCasr6yQ/mqdefault.jpg)

