Two Stage Diffusion Based Text To Image Generation With Clip Embeddings

via “latent-space text-to-image generation with dual-text-encoder architecture”

text-to-image model by undefined. 20,22,003 downloads.

Unique: Dual-text-encoder architecture combining OpenCLIP (semantic understanding) and CLIP (alignment) instead of single CLIP encoder used in SD 1.5, enabling richer semantic grounding; two-stage training pipeline (256→1024) produces native 1024×1024 output without cascading upsampling, reducing artifacts and inference steps vs. prior approaches

vs others: Outperforms Stable Diffusion 1.5 on semantic consistency and resolution quality while maintaining similar inference speed; more accessible than Midjourney/DALL-E 3 (open-source, no API costs) but slower inference than distilled models like LCM-LoRA