vishal
チャンネル登録者数 121人
42 回視聴 ・ 2いいね ・ 2024/12/11
In this video, I implement negative prompting in Stable Diffusion by modifying the basic diffusion loop. Starting with standard classifier-free guidance, I show how to replace the unconditional noise prediction with a negative prompt to steer generation away from undesired attributes. Using examples, I demonstrate how different guidance scales affect the final image.
Colab notebook: colab.research.google.com/drive/1GdTZGkQ-LDdW5hK8t…
Blog post (with additional failed implementations): vishalbakshi.github.io/blog/posts/2024-11-20-negat…
diffusers Implementation: github.com/huggingface/diffusers/blob/43534a8d1fd4…
00:00 - Introduction
02:33 - Encode Text
07:54 - Create Noisy Latents
09:23 - Diffusion Loop
15:57 - Decoding Latents
19:58 - The `mk_samples` function
24:53 - Negative Prompting Implementation
28:13 - Guidance Scale Experimentation
33:06 - diffusers Implementation
コメント
関連動画
19:12
TIL: Understanding LLM Foundry's BinPackCollator (Sequence Packing for 95% Token Efficiency!)
40 回視聴 - 10 日前
22:47
My Second-Place Winning Tiny Model Hackathon Journey: Pre-Training from Scratch
24 回視聴 - 3 週間前
15:06
Building an LLM Judge Agreement App: 7 Iterations from Basic to Full Functionality
30 回視聴 - 2 週間前
26:51
TinyScale Lab: Exploring the Connection Between Training Dynamics and Model Capabilities
12 回視聴 - 3 週間前
24:19
Brooklyn Bridge Timeline of the Allision by the Mexican Naval Training Ship ARM Cuauhtémoc
384,480 回視聴 - 23 時間前
使用したサーバー: wakameme
コメントを取得中...
関連動画
TIL: Understanding LLM Foundry's BinPackCollator (Sequence Packing for 95% Token Efficiency!)
40 回視聴
TinyScale Lab: Exploring the Connection Between Training Dynamics and Model Capabilities
12 回視聴
Brooklyn Bridge Timeline of the Allision by the Mexican Naval Training Ship ARM Cuauhtémoc
38万 回視聴
コメントを取得中...