<AUDCAP> <ENDAUDCAP> broken?
I just tried this model for the first time. Using the following prompt:
"ambient background sound of a busy city at night Photorealistic, hyperrealism. A sleek, metallic, androgynous angel statue, wings spread wide, mid-flight taking off from a neon-lit cyberpunk city. Just before she takes off she turns to the camera and asks: < S>So this is what you wanted?. The statue is coming to life, small sparks of energy emanating from the wings and body as it animates. The city below is densely packed with towering skyscrapers adorned with vibrant, glowing holographic advertisements written in Japanese. The overall lighting is dramatic, with a blend of cool blues and purples from the city lights and warm oranges and yellows from the energy around the statue. The statue's expression is determined, eyes glowing with a soft light. Focus on the statue's dynamic pose, capturing the moment of powerful liftoff. The background should have a soft bokeh effect to emphasize the statue in the foreground. 8k resolution, highly detailed, cinematic lighting."
Everything generated as it should, and I used the preset settings at 50 steps and unipc, the description within the < AUDCAP> < ENDAUDCAP> was just read, and not interpreted as an instruction. Se result below.
im having the same issues
ambient background sound of a busy city at night Photorealistic, hyperrealism. A sleek, metallic, androgynous angel statue, wings spread wide, mid-flight taking off from a neon-lit cyberpunk city. Just before she takes off she turns to the camera and asks: < S>So this is what you wanted?. The statue is coming to life, small sparks of energy emanating from the wings and body as it animates. The city below is densely packed with towering skyscrapers adorned with vibrant, glowing holographic advertisements written in Japanese. The overall lighting is dramatic, with a blend of cool blues and purples from the city lights and warm oranges and yellows from the energy around the statue. The statue's expression is determined, eyes glowing with a soft light. Focus on the statue's dynamic pose, capturing the moment of powerful liftoff. The background should have a soft bokeh effect to emphasize the statue in the foreground. 8k resolution, highly detailed, cinematic lighting.
i suspect it is due to formatting, you need to encapsulate the speech in <S> and <E> with no spaces in the token. also i dont see <AUDCAP> <ENDAUDCAP>
I also noticed that some part of the prompt are talked while being outside <S><E>, but I simply put them after the dialog and it's not talked anymore.