ostris/ideogram_4_turbotime_lora

Hugging Face Models Trending 模型
lora ideogram-4 turbotime few-step image-generation fine-tuning

摘要

一种LoRA，它适配Ideogram 4，可在仅2步且无需CFG的情况下生成高质量图像，采用新颖的连续涡轮训练方法。

任务：text-to-image 标签：diffusers, text-to-image, lora, base_model:ideogram-ai/ideogram-4-fp8, base_model:adapter:ideogram-ai/ideogram-4-fp8, license:other, region:us
查看原文
查看缓存全文

缓存时间: 2026/06/21 04:32

ostris/ideogram_4_turbotime_lora · Hugging Face

来源：https://huggingface.co/ostris/ideogram_4_turbotime_lora

  • Ideogram 4 TurboTime LoRA (https://huggingface.co/ostris/ideogram_4_turbotime_lora#ideogram-4-turbotime-lora)

https://huggingface.co/ostris/ideogram_4_turbotime_lora#ideogram-4-turbotime-loraIdeogram 4 TurboTime LoRA

这个 LoRA 能将 Ideogram 4 变成一个少步数模型。它只需 2 步即可生成高质量图像，无需 CFG 和无需无条件模型。它采用我正在研究并不断完善的一种持续涡轮训练方法创建。待该方法完全打磨成熟后，我会发布更多相关信息。随着方法的改进，该 LoRA 可能会推出更新版本。

ideogram_4_turbo_time (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/hk4OWp6IGuscFhucp458a.jpeg)

以下示例使用默认 AI Toolkit 提示词，步数为 8 步：

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710544679__000021500_0 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/U2LZFNzB-HvAdh18abY6i.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710551105__000021500_1 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/yTwdKZlIs3WPqKQsPJ4dD.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710557512__000021500_2 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/yokPp8KTFU1dEiCY6JiSR.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710563838__000021500_3 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/hCMzvRp-f-V1VkKLr8Jet.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710570261__000021500_4 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/NyCluscttfmUkXie_A1qR.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710576750__000021500_5 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/ZqcHGa8r7OQPkOWwJpBcL.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710583240__000021500_6 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/UWIX2p-vPwKhv3qIIznuS.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710589316__000021500_7 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/76D-LhoU2DV0XhFcOrbuE.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710595579__000021500_8 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/dJBsv8u7gGmhb8eNV-9_u.jpeg)

_mnt_Data_train_out_ui_ideogram_turbo_8_v1_samples_1781710602041__000021500_9 (https://cdn-uploads.huggingface.co/production/uploads/643cb43e6eeb746f5ad81c26/wjPl3Mf5oJVBayCW9XUW7.jpeg)

上月下载量：1,679

模型树 forostris/ideogram_4_turbotime_lorahttps://huggingface.co/docs/hub/model-cards#specifying-a-base-model

使用ostris/ideogram_4_turbotime_lora的空间1

相似文章

ideogram-ai/ideogram-4-nf4

Hugging Face Models Trending

Ideogram 发布了 Ideogram 4，这是他们首个从头训练的开放权重文本生成图像模型，具备业界领先的多语言文字渲染能力、JSON 结构化提示词、边界框布局控制以及原生 2K 分辨率输出。NF4 量化版本已在 Hugging Face 上发布，该模型被认为是目前最优秀的开放权重图像模型，可与专有前沿模型相媲美。

Ideogram 4 (GitHub 仓库)

TLDR AI

Ideogram 4 是一个从零开始训练的开放权重的文生图模型，支持结构化 JSON 提示、业界最佳的多语言文本渲染、边界框布局控制、调色板控制以及原生 2K 分辨率输出。

drbaph/MiniMax-H3-Turbo-Lora-ComfyUI

Hugging Face Models Trending

MiniMax-H3 Turbo 4步音频-视频生成的第三方ComfyUI兼容LoRA转换，包含进一步训练的checkpoint-500变体及示例工作流。

larryvrh/MiniMax-H3-Turbo-Lora

Hugging Face Models Trending

这是 MiniMax-H3 的早期预览版 LoRA，支持在 4 步采样（而非约 20 步）内联合生成视频和同步音频，采样速度约提升 5 倍，并附带 ComfyUI 自定义节点和三个 bf16 检查点。