↑ yükselişteHF trend +945 (1962)görselmodel3D·reddit-stablediffusion·3 saat önce
Hey everyone! In my previous post , I mentioned that alongside Qwen Image 2.1, fine-tuned models called Prompt Enhancers (PE) were released to optimize user prompts. They help refine prompts for standard text-to-image (T2I), image-to-image (I2I), and reference-based generation. Back then, we noticed that using PE for T2I isn't always worth it, as the results often fall short compared to using raw prompts. However, when it comes to I2I , the PE setup truly shines - its ability to write precise instructions and strictly follow the prompt is genuinely impressive. Today, I want to show you what…
↑ yükselişteHF trend +945 (1962)3D·reddit-stablediffusion·2 saat önce
Sharing a simple Qwen Image 2.1 workflow I made for generating character sheets. It can first create a wardrobe sheet from an outfit reference and generate a face, then combines both using the attached mannequin image to control the poses and layout. The face and wardrobe generation parts are optional, so you can just load your own images instead. Some seeds are fixed because they worked well in my tests, but feel free to change or randomize them as needed. You can also replace/edit the mannequin image if you want different poses or a different character-sheet layout…
görselmodel·hf-papers·3 gün önce·25 oy
Vector Quantization (VQ) is fundamental to discrete visual tokenizers that power modern autoregressive and masked image generation models. While recent shared-projection codebook methods have substantially advanced codebook utilization, training stability remains a critical and underexplored challenge. We argue that the root cause lies in the entanglement of the Encoder--Decoder and Codebook training: because neither module can reliably fulfill its own responsibility in isolation, the system can only function when the two subsystems happen to cooperate---a fragile condition that breaks down…