ERNIE-Image (Turbo) on 6 GB of VRAM
Not at full quality. The 16-bit and 8-bit files are too big for 6 GB; the largest that fits is Q2_K (3.2 GB). Baidu's 8B model from April 2026, strong at rendering text in images. Turbo runs in about 8 steps; Base takes 50.
Q2_K (3.2 GB) fits with only 0.5 GB to spare (about 5.5 GB of 6 GB). Quality: heavy loss; a last resort. For better quality, Q3_K_M (3.9 GB) also runs, with about 0.2 GB spilling into system RAM - a little slower, still practical.
01Other amounts of VRAM
02Every file, on 6 GB
| File | Size | Needed | On this card | Quality | Download |
|---|---|---|---|---|---|
| 16-bit SAFETENSORS · Comfy-Org | 16.1 GB | 18.4 GB | 12.4 GB too big | the original weights | Hugging Face → |
| Q8_0 GGUF · unsloth | 8.7 GB | 11.0 GB | 5.0 GB too big | practically identical to the original | Hugging Face → |
| Q6_K GGUF · unsloth | 6.8 GB | 9.1 GB | 3.1 GB too big | very close to the original | Hugging Face → |
| Q5_K_M GGUF · unsloth | 5.9 GB | 8.2 GB | 2.2 GB too big | close; small differences in fine detail | Hugging Face → |
| Q4_K_M GGUF · unsloth | 5.0 GB | 7.3 GB | spills 1.3 GB | good; some loss in fine detail and text | Hugging Face → |
| Q3_K_M GGUF · unsloth | 3.9 GB | 6.2 GB | spills 0.2 GB | noticeable loss of detail | Hugging Face → |
| Q2_K ← GGUF · unsloth | 3.2 GB | 5.5 GB | fits · 0.5 GB spare | heavy loss; a last resort | Hugging Face → |
“Needed” adds 1.5 GB of working memory for ERNIE-Image and a 0.8 GB reserve. Sizes 2026-10-01.
03What to download
| Part | File | Folder | Size | |
|---|---|---|---|---|
| Model | ernie-image-turbo-Q2_K.gguf Q2_K · unsloth/ERNIE-Image-Turbo-GGUF | models/unet | 3.2 GB | Download → |
| Text encoder | ministral-3-3b.safetensors Ministral 3 3B · Comfy-Org/ERNIE-Image | models/text_encoders | 7.7 GB | Download → |
| VAE | flux2-vae.safetensors FLUX.2 VAE · Comfy-Org/ERNIE-Image | models/vae | 0.3 GB | Download → |
| Total download · keep about the same free on disk | 11.2 GB | |||
Only for some workflows: ERNIE prompt enhancer (optional) (6.9 GB). Sizes from Hugging Face, 2026-10-01.
04The text encoder
At 7.7 GB even the smallest Ministral 3 3B file is too big for 6 GB, so it runs from system RAM. That slows down reading the prompt, not the picture. Sizes: 16-bit 7.7 GB (an optional prompt-enhancer LLM (6.9 GB) can be skipped). Why the encoder does not count against VRAM →