Avoiding pitfalls of AI for designers: Guiding principles – LogRocket Blog https://blog.logrocket.com/ai-product-design-guiding-principles/
“Lumina2 LoRA fine-tuning just dropped. Enjoy the Apache 2.0 awesomeness along with other great stuff that Lumina2 delivers in its size! Its system prompting capability through Gemma2 needs further exploration but I have some results ⬇️ https://x.com/RisingSayak/status/1892462411451412674
“Awesome research from ByteDance continues. Current methods of Subject-to-video merges text prompts and reference images to produce consistent videos, yet many approaches fail to preserve subject fidelity. This new research Phantom merges text and reference-image features in a https://x.com/rohanpaul_ai/status/1894000198210490440
“Say hello to Ideogram 2a, our fastest and most affordable text-to-image model to date — optimized for graphic design and photography. Now live on the Ideogram website, API, and partner platforms for all users. https://x.com/ideogram_ai/status/1895157668102222075
“Two European startups @recraftai and @bfl_ml lead image generation in the world. The third place model, Imagen 3, was developed in London but under a Californian company. https://x.com/NandoDF/status/1894337775832334564
XLabs-AI/flux-lora-collection · Hugging Face https://huggingface.co/XLabs-AI/flux-lora-collection
“The human ingenuity behind the Grok logo is something a diffusion model still lags significantly behind, even after grokking trillions of images. Very few curves and parts, but the emotions they evoke are deep and raw. It’s a visual poetry with minimal pixels. 👏” / X https://x.com/DrJimFan/status/1892462786426593392
“Wrote a blog to go through my favorite flavors of attention, commonly seen in the diffusion-based image & video generation models. Includes cross-attn, joint-attn (& friends), and SANA-style linear attn. Also includes common goodies like GQA, parallel-layers, etc. with code. https://x.com/RisingSayak/status/1895066818747998561
“Large Language Diffusion with Masking (LLaDA) are here – and their generation looks so fucking dope! 🤯 True to @ylecun’s vision, Ditch the auto-regressive bits and approximate the language distribution via Maximum Likelihood Estimation! So cool to watch the model denoise text https://x.com/reach_vb/status/1894847408619794493
“FlexTok: Resampling Images into 1D Token Sequences of Flexible Length New paper from Apple and EPFL – “We introduce FlexTok, a tokenizer that projects 2D images into variable-length, ordered 1D token sequences. For example, a 256×256 image can be resampled into anywhere from 1 https://x.com/iScienceLuvr/status/1892550422050877486
