File size: 3,621 Bytes
7b353b1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 | ---
language:
- en
- ru
- zh
base_model:
- ideogram-ai/ideogram-4-fp8
- ideogram-ai/ideogram-4-nf4
tags:
- gguf
- text-to-image
- image-generation
- comfyui
- stable-diffusion.cpp
pipeline_tag: text-to-image
license: other
license_name: ideogram-non-commercial-model-agreement
license_link: https://huggingface.co/ideogram-ai/ideogram-4-fp8/blob/main/LICENSE.md
---
# Ideogram4 GGUF quantized files
```tree
.
├── diffusion/
│ ├── cond/
│ │ ├── ideogram4_Q4_0.gguf
│ │ ├── ideogram4_Q4_1.gguf
│ │ ├── ideogram4-Q4_K.gguf
│ │ ├── ideogram4-Q5_0.gguf
│ │ ├── ideogram4_Q5_1.gguf
│ │ ├── ideogram4_Q5_K.gguf
│ │ ├── ideogram4-Q6_K.gguf
│ │ └── ideogram4-Q8_0.gguf
│ └── uncond/
│ ├── ideogram4_unconditional_Q4_0.gguf
│ ├── ideogram4_unconditional_Q4_1.gguf
│ ├── ideogram4_unconditional_Q4_K.gguf
│ ├── ideogram4_unconditional_Q5_0.gguf
│ ├── ideogram4_unconditional_Q5_1.gguf
│ ├── ideogram4_unconditional_Q5_K.gguf
│ ├── ideogram4_unconditional_Q6_K.gguf
│ └── ideogram4_unconditional-Q8_0.gguf
├── text_encoder/
│ ├── Qwen3-VL-8B-Q4_0.gguf
│ ├── Qwen3-VL-8B-Q4_1.gguf
│ ├── Qwen3-VL-8B-Q4_K_S.gguf
│ ├── Qwen3-VL-8B-Q4_K_M.gguf
│ ├── Qwen3-VL-8B-Q5_K_S.gguf
│ ├── Qwen3-VL-8B-Q5_K_M.gguf
│ ├── Qwen3-VL-8B-Q6_K.gguf
│ └── Qwen3-VL-8B-Q8_0.gguf
└── vae/
│ ├── flux2-vae.safetensors
│ └── flux2-hdr-vae.safetensors
└── lora/
├── realism_engine_v3.safetensors
├── big_boobs.safetensors
├── cum.safetensors
├── innie_vulva_x.safetensors
├── vintage_beauties_womans.safetensors
├── missionary_sex.safetensors
├── 80s_anime.safetensors
├── penis.safetensors
└── penix.safetensors
```
### Model Selection & Quantization Guide
To balance generation quality, memory usage, and inference speed, we recommend the following quantization choices for each component:
#### 1. Conditional Diffusion Model (`diffusion/cond/`)
* **Recommended:** `Q6_K` or `Q8_0`
* Since this model handles the main conditional generation pass, keeping a higher quantization level is key to preserving detail and prompt adherence.
#### 2. Unconditional Diffusion Model (`diffusion/uncond/`)
* **Recommended:** `Q4_K` or `Q5_K`
* **Note:** Using `Q6_K` or `Q8_0` for the unconditional model is generally unnecessary (overkill) and may slow down generation without providing a noticeable improvement in quality.
#### 3. Text Encoder (`text_encoder/`)
* **Recommended:** `Q5_K_M` or `Q4_K_M`
* These medium-sized "K-measure" quants offer a good trade-off, retaining the text encoder's comprehension capabilities while fitting within reasonable memory limits.
---
### General Recommendations for Quantization Types
If you are optimizing for inference speed or trying to fit a specific model entirely into VRAM/RAM, keep these rules of thumb in mind:
* **Prefer `_K` variants over `_0` and `_1`:** When choosing between `Q4` or `Q5` options, always prefer the `_K` variants (e.g., `Q4_K_M`, `Q5_K_M`, or standard `_K`).
* **Avoid `_0` and `_1` if possible:** The older `_0` and `_1` quants (like `Q4_0` or `Q4_1`) perform worse in terms of quality loss. While they are marginally smaller, the minor size reduction rarely justifies the drop in generation quality compared to `_K` equivalents. |