📦 This LoRA is from Muinez on Hugging Face. Download it using the link below and import it into Draw Things. ⚙️ Set the LoRA weight to 1 and use 4 steps. 🎨 Recommended sampler: Euler a trailing. 🧠 Base model: Flux.3 [Klein] 5.6B Base. 🔗 Download link: huggingface.co/Muinez/flux3-…

🤯 THIS IS ABSOLUTELY INCREDIBLE! ⚡ 15s(768*1024) Flux.3 Klein Base generation with amazing quality on an M5 iPad Pro! 🚀 All you need to do is add this 4-step Distill LoRA and set the shift to 3! 👇 Go try it out on Draw Things! Details below!

Just switch your system appearance to Dark Mode in Settings, and the Draw Things app will automatically switch to a dark interface too.

🚀 Draw Things v26.1020.0 is officially live on the App Store! 💥 FLUX.3 [klein] 5.6B series models are now fully supported. 🌿 MOSS Image series models join the lineup — fresh and ready to explore. 🔧 Several crashes and UI issues have been squashed. 📱 Update now and put these new models to work.

🚨 These two frontier models are coming to Draw Things. Stay tuned!

🚨 Draw Things rolled out some updates recently （v26.1008 & v26.1013）—we reworked the Moodboard UI and layout to make your workflow much faster! 1️⃣ New Moodboard UI: allows extracting depth maps directly, and now we hide the "Control" tab for models that don't support it （switch back to SDXL and you'll see the Control tab reappear）. 2️⃣ Silently updated Qwen Image 2.1 VAE to the madebyollin variant （fixes checkerboard artifacts and numerical overflow under fp16）. 3️⃣ General speed improvement on M5 Ultra; allow ANE+GPU to run at the same time, bringing noticeable improvements on 8-bit S models across M3 Ultra /M4 Pro / M4 Max, and even bigger gains on larger models such as Krea 2 Turbo. 4️⃣ Fixed some crashes, and resolved Krea 2 generation issues (actually, it is the VAE decoder) at 2048x2048 resolution. 5️⃣ gRPCServerCLI and draw-things-cli updated to 26.1014.0. gRPCServerCLI on CUDA also supports 4-bit S format properly with this release. 6️⃣ We are raising Draw Things+ subscription to $12.99 per month to help secure more hardware such that some more desirable models will be possible to offload to. It will be in effect on November 7th, 2026.

After another 30 minutes.

DeepSeek 4.1 Flash 0910 from @localcodeapp insists that the bar is not correct, and spent another hour to refine that. This bar chart is accurate now.

In the past a few days, we ran benchmarks across 4 inference engines, on 4 Macs, with 3 community-beloved models, you can read more about this in releases.drawthings.ai/p/har…

🔗 Discord community post with download link: discord.com/channels/1038516… 🌐 Download from the developer’s personal blog: qxm.de/dt-helper-en#dtinspec…

🔥 The Draw Things community is thriving, with passionate developers and enthusiasts building amazing tools! 🛠️ More and more fantastic community-made tools keep emerging. 🗂️ DTInspector makes it easy to manage your Draw Things output folder. 🎁 Lightweight, clean, and completely free! 🙏 Huge thanks to our Discord community member ‘Moof’ for this excellent work. 👇 Check out the download details below!

Copy and paste it into your Draw Things app and share your feedback! (Of course, the settings below use Qwen Image 2.1 4-bit S. You can switch to whichever quantization format you prefer.) —————— {"refinerModel":"","batchSize":1,"maskBlurOutset":0,"seed":654750154,"sampler":14,"maskBlur":1.5,"batchCount":1,"seedMode":2,"faceRestoration":"","width":832,"sharpness":0,"strength":1,"preserveOriginalAfterInpaint":false,"colorCalibration":"none","resolutionDependentShift":true,"tiledDiffusion":false,"guidanceScale":1,"controls":[],"steps":10,"model":"qwen_image_2.1_i4x.ckpt","tiledDecoding":false,"cfgZeroStar":false,"upscaler":"","hiresFix":false,"loras":[],"shift":1.9863333000000001,"height":1216} ——————

🚨 No acceleration LoRAs needed for Qwen Image 2.1 in Draw Things! ⚙️ Try this optimal sampler settings for both image generation and editing tasks. ⚡ Achieve fast, high-quality outputs in just 6–10 steps without any speedup LoRAs. 🎨 Enjoy pure raw base model quality, completely free from LoRA artifacts and distortion. 👇 Check out the detailed Draw Things configuration parameters below!

Of course.

Correct. Official Models only display free license models (free as in freedom).

🚨 Qwen Image 2.1, the compact, efficient, and unified image creation model, is now officially supported in Draw Things! You need this amazing model! 🔄 Draw Things v26.0924.0 was released on the iOS / macOS App Store one hour ago. This version brings: 🔹 Added Qwen Image 2.1 support, including LoRA import. A 4-bit S version is also now available. 🔹 Added Sol Attention support to speed up MiniMax H3. 🔹 Some minor UI fixes.

Below are the original Hugging Face repository for TaoMate-H3 and the Kijai version I tested in Draw Things. The Kijai version is 182 MB and performs quite well in my tests. The video above was also generated using this version. Try it and share your feedback! 🔗TaoMate-H3: huggingface.co/TaoLiveAIGC/T… 🔗TaoMate 3-Step LoRA / Kijai: huggingface.co/Kijai/MiniMax…

🚨 Generating MiniMax H3 Videos in Just 3 Steps! 🧪 The TaoMate-H3 3-Step LoRA is definitely worth trying. 🚀 Import it into Draw Things and the results are incredibly impressive. 👇 Click below and check the comments for more details.

💡 A few things you should know about MiniMax H3 in Draw Things! 🔷 MiniMax H3 is the first model in Draw Things to support first-and-last-frame generation. The Canvas is used as the first frame, while you can place an image in the Moodboard as the last frame. It supports T2VA / I2VA / FL2VA. 🔷 MiniMax-H3 [ref2va] can be used for reference-based generation. Provide reference images through the Canvas and Moodboard, and identify them in the prompt as <Picture 1>, <Picture 2>, etc. 🔷 Currently, Draw Things does not support using Audio as a reference input for video generation, nor does it support L2VA (last-frame instruction) generation.