Draw Things

🚨 We are welcoming a MAJOR UPDATE! 🔄 Draw Things v1.20260314.0 was released on the iOS / macOS App Store today. This version brings: 🔹Support for LTX-2.3 22B series models, including spatial upscalers. 🔹Support for FLUX.2 [klein] 9B KV for faster editing capability. 🔹Improved Video VAE performance for various video models. gRPCServerCLI updated to 1.20260314.0 with the above improvements.

👉 When you use LTX-2.3 in Draw Things to create a cinematic clip inspired by famous Chinese classical poetry…

@dadabots Probably not us. We don't use MLX, and LTX-2.x are the first models we support w/ audio.

@drawthingsapp love it! been drawing inspiration from your MLX code w/ claudecode for stable audio

💥 When AI starts writing high-performance Metal shaders, everything changes! 🚀 LTX-2.3 VAE decoding is boosting. 🧪 Go try it yourself in the latest version of Draw Things.

LTX 2.3 22B Distilled M5 Max 40 GPU cores WINS vs M3 Ultra 80 GPU cores in generation of a 5 seconds video: 🥇 M5 Max 121 secs 🥈 M3 Ultra 206 secs I have used again Draw Things App for this test with default settings. Guess which video is from which machine, Audio on!

3️⃣ This is the clip where we enabled High-Res Fix to make the Spatial Upscaler work, with a resolution of 1152×768 and a first pass size of 576×384. ⚙️ The parameters are as follows (you can copy and paste for reference). —————— {"hiresFixWidth":576,"width":1152,"seed":613921383,"loras":[],"cfgZeroStar":false,"steps":8,"compressionArtifactsQuality":70.5,"preserveOriginalAfterInpaint":true,"faceRestoration":"","sharpness":0,"hiresFixHeight":384,"refinerModel":"","shift":7.8200000000000003,"cfgZeroInitSteps":0,"tiledDiffusion":false,"batchSize":1,"upscaler":"","hiresFix":true,"sampler":10,"height":768,"maskBlurOutset":0,"controls":[],"compressionArtifacts":"disabled","numFrames":169,"model":"ltx_2.3_22b_distilled_q6p.ckpt","seedMode":2,"maskBlur":1.5,"strength":1,"batchCount":1,"causalInferencePad":0,"tiledDecoding":false,"hiresFixStrength":0.69999999999999996,"guidanceScale":1} ——————

The base image on the canvas is this, and the prompt and config are as follows: —————— 1️⃣Prompt：flowing lines move smoothly across the frame as the camera slowly rotates, creating a dynamic and high-end visual rhythm. The words LTX-2.3 stays centered, reflecting textured light and subtle highlights as lighting shifts across its surface. A deep, powerful voice says, “LTX-2.3 is now on Draw Things, Go try it now!” accompanied by a soft metallic sweep sound. —————— 2️⃣Config： —————— {"causalInferencePad":0,"controls":[],"seed":1316453444,"batchCount":1,"cfgZeroInitSteps":0,"loras":[],"maskBlur":1.5,"compressionArtifacts":"disabled","tiledDiffusion":false,"sampler":10,"strength":1,"numFrames":177,"sharpness":0,"shift":7.8200000000000003,"steps":8,"compressionArtifactsQuality":70.5,"seedMode":2,"preserveOriginalAfterInpaint":true,"model":"ltx_2.3_22b_distilled_q6p.ckpt","batchSize":1,"maskBlurOutset":0,"hiresFix":false,"cfgZeroStar":false,"tiledDecoding":false,"height":512,"upscaler":"","faceRestoration":"","width":768,"guidanceScale":1,"refinerModel":""} ——————

🔊 Image to Audio-Video！ 💥 LTX-2.3 is now on Draw Things, Go Try it Now！ 📝 Settings and Prompt for this example In the Comments！

@ivanfioravanti @daveyjones9091 Try 1.20260314.0, hot off the press: engineering.drawthings.ai/p/2-days-to-sh…

@daveyjones9091 I am right now!!! With @drawthingsapp Stay tuned!

⚡️ What a BOOSTER for multi-image editing! 🔥 FLUX.2 Klein 9B KV-Cache is a total banger. We'll be bringing this to you guys in Draw Things ASAP!

🍳Cooking up some great new features...

⬇️Configuration below (Copy & Paste if you like): ———————— {"batchSize":1,"loras":[],"tiledDiffusion":false,"model":"ltx_2_19b_distilled_q8p.ckpt","maskBlurOutset":0,"hiresFix":false,"causalInferencePad":0,"batchCount":1,"shift":5,"guidanceScale":1,"preserveOriginalAfterInpaint":true,"maskBlur":1.5,"numFrames":161,"seedMode":2,"seed":412881376,"strength":1,"compressionArtifacts":"disabled","cfgZeroStar":false,"controls":[],"sharpness":0,"steps":9,"width":768,"compressionArtifactsQuality":43.100000000000001,"cfgZeroInitSteps":0,"sampler":10,"refinerModel":"","upscaler":"","faceRestoration":"","height":512,"tiledDecoding":false} ————————

🎬 Sharing some 161-frame（near ~7s）dialogue scenes using LTX-2 distilled! 🧪 All rendered in Draw Things. 👇 I've posted the parameters in the thread below for your reference!

🎬 The four short videos above were all generated using the same parameters. 👇🏻 For reference, the corresponding prompt is shown in the image below: —————— {"seed":2157389691,"numFrames":121,"faceRestoration":"","shift":11,"tiledDecoding":false,"steps":8,"cfgZeroInitSteps":0,"tiledDiffusion":false,"maskBlur":1.5,"upscaler":"","strength":1,"causalInferencePad":0,"sampler":15,"controls":[],"maskBlurOutset":0,"model":"ltx_2_19b_distilled_q8p.ckpt","guidanceScale":1,"batchSize":1,"loras":[],"preserveOriginalAfterInpaint":true,"refinerModel":"","sharpness":0,"batchCount":1,"seedMode":2,"compressionArtifacts":"disabled","compressionArtifactsQuality":43.100000000000001,"height":448,"cfgZeroStar":false,"width":832,"hiresFix":false} ——————

🎬 Did you create any audio-video with LTX-2 in Draw Things today? 🧪 The following examples were generated using these settings. 📌 Parameters are in the comments. Go check it out, hope it helps!

👇🏻The LoRA link is here: huggingface.co/alibaba-pai/Z-… Note: I have already placed the Draw Things image generation parameters for the two LoRAs in the ALT information of the image above, which you can use as a starting reference for generating images.

🚀 Run Z-Image in Just 2 Steps, Really Impressive！ 🔥 The latest Z-Image-Fun-Lora-Distill-2-Steps-2603 and 4-Steps-2603 LoRAs are TRULY bangers! ✨ Thanks to the addition of a random timesteps strategy, the results this time are absolutely stunning. 💎 While the 2-Steps LoRA delivers solid results, the 4-Steps LoRA can only be described as exceptional.

🚑 We have submitted two more updates to the App Store with several hotfixes. Please take note — the latest version is Draw Things v1.20260304.0: → Fix metadata for LTX-2 models. → Fix Z-Image LORA trained by OneTrainer import issues. → Revert to index to the last frame for generated videos to facilitate some existing scripts workflow. → Fix Compute Units calculation issues for FLUX.1. → Fixed a crash when generating LTX-2 video with “Save Generated Media to” enabled. → New Compute Units formulation that encourages video generation with Cloud Compute. → New audio playback UI for video with audio.

6️⃣ The Last One——Image To Video！ This guy in the sports jersey looks directly into the lens with a confident, steady gaze. He speaks with subtle lip movements, saying "The louder the crowd, the quieter my mind." The camera performs a slow, dramatic zoom toward his face while the background city lights flicker in soft bokeh. Cool blue moonlight and sharp rim lighting define his silhouette against the misty night air.

5️⃣Medium close-up frontal shot of a young girl centered in frame, captured with a shallow depth of field and a soft cinematic composition. Warm ambient lighting creates gentle highlights on her face and smooth skin tones, with a softly blurred background in pastel colors and subtle bokeh. She raises both hands in front of her chest, slowly forming a heart shape with her fingers, then leans slightly toward the camera while maintaining steady eye contact. Her lips curve into a gentle smile, her eyes slightly narrowed with relaxed eyelids, and her shoulders remain calm and still. The camera performs a slow forward dolly push, transitioning from a medium close-up to a tighter close-up, emphasizing her facial expression and hand gesture. Soft ambient room tone blends with warm, slow-tempo background music, creating a tender atmosphere. She speaks in a soft, affectionate voice, “I love you.”

📼 Some LTX-2 video examples and prompts generated in Draw Things: Click for more👇🏻 1️⃣ A young male programmer wearing a baseball cap sits at his desk in a medium close-up, lit by the cool glow of his monitor in a dim room. His voice comes in immediately with excited energy over soft keyboard clicks as he says, “Life is boring, let’s coding.” Blue screen light reflects in his eyes while he leans toward the computer, then turns slightly to the camera with raised eyebrows and a bright grin. The camera makes a gentle forward push as the steady typing sound continues in the background.