LoRA

Timeline

1. Oct 19, 2026 · Opinion / analysis · 1 source
   How good is FLUX.3 Klein at editing? A 12-task comparison that surprised me
   All we've been hearing for the last few days is how FLUX.3 Klein didn't live up to expectations when it comes to text-to-image.
2. Oct 19, 2026 · Tutorial / explainer · 1 source
   Fixed the vertical colour stripes in MOSS-Image SFT with a 1.9 MB LoRA (extracted, not trained)
   If you've used MOSS-Image SFT you may have noticed faint vertical colour stripes in skies, fog, walls and studio backdrops.
3. Oct 18, 2026 · Opinion / analysis · 1 source
   Help identifying anime eye style + which model/prompt gets this look?
   I've been looking for this specific anime art style for months, and I'd love to be able to generate it myself.
4. Oct 18, 2026 · Tutorial / explainer · 1 source
   [P] asmsense: 228k stripped x86-64 functions with source-aware labels, and a map of the data we still lack. [P]
   asmsense is a dataset and a small model (Qwen3-8B + LoRA) for naming and describing functions in stripped binaries.
5. Oct 18, 2026 · Opinion / analysis · 1 source
   Same prompt, 3 models: SDXL (WAI-Illustrious v17) vs Krea 2 turbo vs MOSS-Image-SFT (anime scenes, no LoRA)
   Tested the three side by side using the exact same text prompt (booru tags only, no style prefix, no quality tags, no negative prompt), 4 scenes × 2 seeds, 1024×1024.
6. Oct 18, 2026 · Opinion / analysis · 1 source
   FastH3 and LoRAs
   But when I connect the model output to Basic Guide and Basic Scheduler, two nodes model outputs (Model Attention Backend and Model Sparse Attention) lose the connection to the Basic Guider and Basic Scheduler which makes my FastH3... well, not fast.
7. Oct 18, 2026 · Opinion / analysis · 1 source
   Trained a Qwen-Image-2.1 LoRA with fewer restrictions and omni poses
   Been experimenting with fine-tuning Qwen-Image-2.1 and finally got around to releasing the LoRA.
8. Oct 18, 2026 · Opinion / analysis · 1 source
   MOSS-Image comes to Fizgig
   Fizgig trains on MOSS-Image Base, the checkpoint OpenMOSS recommend for LoRAs, and previews on MOSS-Image SFT, the one most people generate with.
9. Oct 18, 2026 · Model release · 1 source
   Sharing my ultimate krea 2 Asian Character Lora
   Unfortunately I don't have any plan to create any new character krea 2 in near future (since krea 3 is very near to be released as open weights, but I might update the old krea 2 lora to refine the dataset when I see some bad seed/result.
10. Oct 17, 2026 · Opinion / analysis · 1 source
    DeepSeek V4.1 Flash 1017 [GGUF + Abliterated]
    Hi all, as many of you are probably aware DeepSeek V4.1 Flash 1017 was released yesterday.
11. Oct 17, 2026 · Opinion / analysis · 1 source
    Vertical Banding in MOSS-Image experiments
    Been digging into the vertical banding in MOSS-Image SFT.
12. Oct 17, 2026 · Opinion / analysis · 1 source
    MOSS-Image extension for Forge Neo
    I've been working on a new extension that brings MOSS-Image to Forge Neo.
13. Oct 17, 2026 · Opinion / analysis · 1 source
    Blown away by Minimax H3 sometimes. It nailed this one!!!!
    • The song, made with Suno, lyrics created by Claude.
14. Oct 17, 2026 · Opinion / analysis · 1 source
    MOSS-Image LoRA training is now in AcademiaSD LoRAlab Trainer Studio (Base and SFT, under 6 GB VRAM)
    AcademiaSD LoRAlab Trainer Studio now supports MOSS-Image, the new open text-to-image model (Apache 2.0).
15. Oct 17, 2026 · Opinion / analysis · 1 source
    Fizgig Lora Isolation - A new process for more powerful LoRAs
    Fizgig LoRA Isolation: train 2 identities into one LoRA and keep everyone else looking like themselves.
16. Oct 17, 2026 · Opinion / analysis · 1 source
    MossImage quants & SFT fix
    I recommend the INT8 as a start. nVidia RTX20x+ & RDNA2 (RX6000)+ are good.
17. Oct 16, 2026 · Opinion / analysis · 1 source
    MOSS-Image (5B, Apache 2.0) on a 12 GB RTX 3060: INT8 quant + lodestones' 8-step DMD turbo in ComfyUI - 35 hard prompts, one seed, no cherry-picking
    Tested the new MOSS-Image (5B single-stream DiT, Qwen3.5-9B text encoder, FLUX.2 VAE, Apache 2.0) on a plain RTX 3060 12 GB / 16 GB RAM with 35 deliberately hard prompts: dense on-image text in English/Chinese/Japanese, accents, exact counts, attribute binding, spatial relations, negation, hands, clocks, very long prompts, and MOSS's Layout JSON.
18. Oct 16, 2026 · Open-source release · 1 source
    axolotl-ai-cloud/axolotl v0.21.0
    This release is headlined by Reflection's Beam, a 501B-parameter MoE you can QLoRA-tune on a single node.
19. Oct 16, 2026 · Research paper · 1 source
    How Finite-Rank Bottlenecks Shape the Low-Rank Adaptation Landscape
    Low-rank adaptation (LoRA) has become the standard method for parameter-efficient fine-tuning of large pretrained models, yet theoretical explanations for why low-rank updates suffice remain incomplete.
20. Oct 16, 2026 · Research paper · 1 source
    Kingfisher: Fast Personalized Visual Autoregressive Generation via Scale-Aware Prompt-Conditioned Scale Decomposition
    Based on this insight, we introduce Kingfisher, a model-tuning-free framework that decomposes early residual scale codes into content- and style-related components through prompt-conditioned scale-code subspaces.
21. Oct 16, 2026 · Research paper · 1 source
    MAS-Bai: Linguistically Informed Adaptive Multi-Agent Translation for Bowen-Script Bai
    We propose MAS-Bai, a linguistically informed adaptive multi-agent framework for Bowen-script Bai-Chinese translation.
22. Oct 16, 2026 · Research paper · 1 source
    DICL-Qwen3VL: Decoupled Confidence Calibration for Interpretable Workwear Detection
    We propose DICL-Qwen3VL, a decoupled interpretable VLM that integrates structured workwear detection, evidence generation, and separate visual and reasoning confidence estimates.
23. Oct 16, 2026 · Opinion / analysis · 1 source
    Krea2 Realism is back — this time as a full checkpoint! And yes, things got a little wilder
    I've been working on the next release, and this time it's a full checkpoint: a merge of my custom Realism Checkpoint and V3 LoRA.
24. Oct 16, 2026 · Research paper · 1 source
    Competence-Based Evaluation of Logical Reasoning in Large Language Models
    Recent advances in large language models (LLMs) have brought impressive performance across diverse tasks; however, shortcomings remain in logical reasoning, including inconsistencies under linguistic variation.
25. Oct 15, 2026 · Open-source release · 1 source
    Comfy-Org/ComfyUI v0.41.0
26. Oct 15, 2026 · Research paper · 1 source
    EgoP$^{2}$C: Internalizing Plan-Perceive-Compose for Spatial Reasoning in Egocentric Video
    Answering complex spatial reasoning questions from egocentric video requires determining what evidence is needed, acquiring targeted observations, and combining them through multiple reasoning steps.
27. Oct 15, 2026 · Model release · 2 sources
    Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint
    Specifically, Granite-Switch-4.1-8B and -30B models are configured to include:
28. Oct 15, 2026 · Open-source release · 1 source
    unslothai/unsloth v0.1.908-beta: NVIDIA Nemotron 3.5 + Meta Muse Spark Open
    Run NVIDIA's new Nemotron 3.5 models and Meta's new Muse Spark Open locally in Unsloth.
29. Oct 15, 2026 · Research paper · 1 source
    Don't Retrain, Remix: Adaptation of Web Agents using Mixture Hypernetworks
    We give web agents a similar ability with WebMix, a mixture hypernetwork that adapts them to unseen interfaces by composing previously trained low-rank adapters.
30. Oct 15, 2026 · Research paper · 1 source
    alkhder at AlexandriaX-2026: Context-Aware English-to-Dialectal Arabic Dialogue Translation with QLoRA-Fine-Tuned NileChat-3B
    This paper describes team Alkhder's submission to Subtask 1 of the AlexandriaX-2026 shared task, which asks systems to translate English dialogue turns into thirteen Arabic dialects.

Often appears with