📜 LDM BERT Text Encode

Encodes a text prompt into a conditioning tensor for the UNet. Uses the BERT-base-uncased tokenizer (77 tokens max). The encoder is a 64-layer transformer with 1280-dim hidden states and 8 attention heads. Prompting tips for this model: • Keep prompts short and concrete (BERT has limited compositional understanding). • Lowercase is fine — the tokenizer is uncased. • Avoid complex syntax, negations, or long narratives. • Single-subject descriptions work best: 'a red car', 'mountain landscape'. • This model was trained without classifier-free guidance — there is no negative prompt.