Version History
Here you can find the changelog of Draw Things: AI Generation since it was posted on our website on 2023-06-20.
The latest version is 1.20240416.0 and it was updated on soft112.com on 19 April, 2024.
See below the changes in each version:
version 1.20240416.0
posted on 2024-04-17
Apr 17, 2024
Version 1.20240416.0
* This fixed a bug where using Tiled Diffusion with CFG > 1 or no-accelerated LoRA can generate undesirable images.
version 1.20240321.2
posted on 2024-03-22
Mar 22, 2024
Version 1.20240321.2
* Fixed memory usage issues when train LoRAs. Now it is possible to train 1024x1024 resolution SDXL LoRA on 16GiB MacBook Pro / Air.
* Fixed HTTP server doesn't do proper conversion of the decoding tile width / height.
* Fixed when "Optimize for Compact Store", we didn't delete old -tensordata file.
version 1.20240215.2
posted on 2024-02-16
Feb 16, 2024
Version 1.20240215.2
* Updated onboarding flow, now it automatically downloads missing files even if you imported models locally;
* Fixed issues with Depth model / captioning model that doesn't work without Flash Attention;
* Prefer internal storage over external storage if a model presented in both places;
* Fixed how we generate random numbers on Metal, now it doesn't have slow memory leak before;
* When text guidance set to 1.0, now all samplers will be twice as fast. This is particularly useful for imported SDXL Turbo-merged models such as TurboVision XL.
version 1.20240212.2
posted on 2024-02-13
Feb 13, 2024
Version 1.20240212.2
* Added Moondream model as image interrogation model. You can customize system prompt to describe image in whatever way you want;
* Updated how we use Depth Anything model for depth map extraction. Now it extracts depth map up to 1036x1036 resolution;
* Updated a new IS Net model for foreground / background extraction;
* Added ControlNet LoRA support;
* Support copy / save extracted depth map to Photos / File for other apps to use.
version 1.20240205.0
posted on 2024-02-06
Feb 6, 2024
Version 1.20240205.0
Fixed a issue that the magic wand mask is never cleared.
version 1.20240128.0
posted on 2024-01-29
Jan 29, 2024
Version 1.20240128.0
* Added "Mask Blur Outset" to help soften inpainting seams along the edge.
* Added "Sharpness" to help improve image sharpness for SDXL generation.
* Fixed various UI related issues.
* Fixed diffusers LoRA import for wider compatibility.
version 1.20240118.2
posted on 2024-01-19
Jan 19, 2024
Version 1.20240118.2
* Added Fooocus Inpaint as SDXL model as well as SDXL LoRA. This is IMHO the best inpaint model for SDXL.
* For recent colors in Paintbrush, we only persist them when it is used now.
* Fixed issue with merging SDXL models.
* Fixed a bug on using multiple LoRAs with 8-bit model. It now generates much better image with LCM LoRA and other LoRAs combined.
version 1.20240114.1
posted on 2024-01-16
Jan 16, 2024
Version 1.20240114.1
* Allow delete projects from the app.
* Disable most actions in the left side bar when image is generating.
* Can trigger image generation with CMD + Enter keyboard shortcut.
* Persist the last 20 selected colors for paintbrush.
* Fixed some v2.1 model importing issues.
* Allow customize Advanced tab with editing advanced_sections.json file.
version 1.20231212.0
posted on 2023-12-13
Dec 13, 2023
Version 1.20231212.0
Addressed some performance issues on Stable Video Diffusion model. Now it should be 10% to 150% faster depending on devices (M2 Pro to M2 Ultra).
version 1.20231204.0
posted on 2023-12-05
Dec 5, 2023
Version 1.20231204.0
* Added SDXL Turbo model. Note that this model is for non-commercial / academic use only.
* Added "Euler A Substep / DPM++ SDE Substep / LCM" as sampler options. The first two are particularly useful for SDXL Turbo models.
* Fixed crashes when using Refiner with the new samplers.
version 1.20231119.0
posted on 2023-11-20
Nov 20, 2023
Version 1.20231119.0
* Fixed crash when importing SD v1 models;
* Added support for LoHa. That means we have full support for LyCORIS now;
* Added IP-Adapter Full Face SD v1.5, IP-Adapter Plus Face SDXL models.
version 1.20231115.0
posted on 2023-11-16
Nov 16, 2023
Version 1.20231115.0
* Added LCM family of models and LoRAs. You can combine LCM family of models / LoRAs to generate high-quality images at as low as 4 steps. On iPhone 15 Pro, that means SD v1 image generation takes less than 10s (at 512x512), SDXL generation takes less than 20s (at 768x768).
* Added temporary fix for performance issues with A17 Pro / M3 chips. Now M3 Pro MBP should be about 20% faster than M1 Pro MBP at closest GPU core count (14-core v.s. 18-core).
* Added SSD-1B family models (LCM SSD-1B) which is about 100% faster than SDXL at reduced quality.
* Speed up image interrogation (image captioning) with KV cache. Caption generation should be 50% to 100% faster now.
* Added experimental support to train LoRAs on top of inpainting models.
version 1.20231104.0
posted on 2023-11-06
Nov 6, 2023
Version 1.20231104.0
* Add metadata so A1111 can recognize the version of the exported LoRA safetensors;
* Add Midjourney prompt migrator to help configure the parameters from Midjourney command-line prompt;
* Add HED preprocessor for softedge ControlNet.
* Remove the titlebar from the UI so it is more efficient use of vertical space.
version 1.20231031.0
posted on 2023-11-01
Nov 1, 2023
Version 1.20231031.0
* Add ability to train T.I. embedding along with the LoRA training.
* Allow to export a T.I. embedding back to safetensors.
* Add support to import SDXL T.I. embeddings.
* Fixes some crashes.
* QoL improvements: sorting model / LoRAs now in Finder order (alphanumerics), long press / right click in training UI data preview to recaption the image.
version 1.20231018.0
posted on 2023-10-19
Oct 19, 2023
Version 1.20231018.0
1. Fixed overcooked loss issue during LoRA training. Now you can use larger learning rate and train longer to improve likeness to your training dataset.
2. Added BLIP2 captioner to generate captions for LoRA training data. Also uses that for the new "Interrogate" option that can generate text description to help with inpainting tasks.
3. Added option to sorting model list / LoRA model list to help discovery.
version 1.20231008.0
posted on 2023-10-09
Oct 9, 2023
Version 1.20231008.0
* Prompt template now can add your own prompt from the JSON file.
* LoRA training can load caption file.
* Fix a bug where 8-bit LoRA cannot be merge properly.
* Fix a bug where we require Metal FlashAttention for training.
* Fix Intel Mac on macOS 14 cannot generate image correctly.
* Some other crash fixes.
version 1.20230915.1
posted on 2023-09-17
Sep 17, 2023
Version 1.20230915.1
This version fixed crashes on Intel Macs.
version 1.20230901.1
posted on 2023-09-03
Sep 3, 2023
Version 1.20230901.1
* Added "moodboard" which you can use with ControlNet Shuffle to provide style guidance to the generation from multiple images.
* Making the default UI font on macOS larger and you can select font size of the interface from Machine Settings.
* Fixed `readEntries` API for script now it can take folder parameter.
* Separated LoRA UI into list of LoRAs and import LoRAs.
version 1.20230826.0
posted on 2023-08-28
Aug 28, 2023
Version 1.20230826.0
* Support "Scripts" which let you automate some tasks such as loading image, moving canvas around, save image on current canvas. As an example, we provided "Tiled Upscale" implementation (a.k.a. "SD Ultimate Upscaler") to show-case such capability.
* Added "Select Multiple ..." in history view such that you can select multiple images in history and delete them at once. Right-click any image in opened history view (with the 4-square button) and select "Select Multiple ..." option.
* Minor fixes on the HTTP API (added sdapi/v1/options endpoint, canvas won't switch sizes back / forth during multi-batch generation).
version 1.20230818.0
posted on 2023-08-18
Aug 18, 2023
Version 1.20230818.0
Better gating for MFA and fixed issues on macOS 13.1 / 13.2.
Added "Negative Original Image Size" for SDXL models which seems to help generated quality.
Added "HTTP API Server" option, which will launch a HTTP API server that you can talk to.
version 1.20230811.0
posted on 2023-08-11
Aug 11, 2023
Version 1.20230811.0
Fixed crashes related to Metal FlashAttention and SDXL model.
version 1.20230801.0
posted on 2023-08-02
Aug 2, 2023
Version 1.20230801.0
* Gated MFA to only enable on non-Intel Macs.
* Fixed issues with importing SDXL model. Now these model can be imported / converted to 8-bit to save disk too.
* Fixed issues with import SDXL LoRAs. Note that there are two types of LoRA: one for SDXL Base and one for SDXL Refiner. Both are supported but they are not compatible with each other.
* Updated vae used for SDXL. You should tap SDXL model again to download the updated vae.
version 1.20230630.0
posted on 2023-07-03
Jul 3, 2023
Version 1.20230630.0
* Introduce 8-bit models. These are the models quantized to 8-bit, but should have very similar output as the original. These models are around 800MiB in size, about 2x smaller than original. You can create 8-bit model off your existing model to save space.
* Fixed crashes related to saving infinite canvas. Now it should work for much larger canvas (we tried 20,000 in width).
* When load photos, it tries the best to load at its native resolution so you can zoom in, fix, and save again, while retaining the original resolution.
version 1.20230625.1
posted on 2023-06-26
Jun 26, 2023
Version 1.20230625.1
* Remember your last selection on save to photos, files or copy to clipboard.
* Ask for output resolution if it deemed too large.
* Fixed a crash related to infinite canvas.
* Added "new seed on generation" option for seed.
version 1.20230615.1
posted on 2023-06-16
Jun 16, 2023
Version 1.20230615.1
Introducing Infinite Canvas! Infinite Canvas allows you to generate images on a infinite size canvas. It allows you to zoom in, inpainting details, fix faces, or zoom out, and generative fill the environment. All your favorite ControlNet, T2I adapters, img2img, LoRA, Textual Inversion, face restoration, upscaler, can be applied seamlessly on the canvas. It truly streamlines the workflow to create any image from your imagination.
Updated new user experience! As a first-time user, you are now presented with 3 options: select a model to download from our official server; import a locally downloaded PyTorch model; or enter a Civitai model URL to download and import model from their server.
Added ControlNet Tile model. ControlNet Tile is best companion to upscale with Infinite Canvas. It helps to retain some semblance even at high denoising strength while adding a lot details.
Added "Paste Civitai Generation Data" option in the context menu of the prompt text box. It parses the model / sampler / steps and other configuration options, while also insert automatic line breaks for long prompt. This feature helps to reproduce sample images from the site.