The Cowart project has garnered over 2,000 stars for its precise AI image editing capabilities using arrow annotations.
Today I'm sharing some tools suitable for image editing. I usually use AI image generators; you can search for styles you like, and there are ready-made suggestions. If you're not satisfied with the result, just generate it a few more times. The real hassle is when the image is already 80-90% satisfactory, but then...
Today I'm posting some things that are suitable for making images.
Normal use AI These are unedited photos; you can search for your preferred style.Prompt wordsThere are also ready-made references; if you are not satisfied with the results, you can generate the code a few more times.
The real problem is that the final image is already 80-90% satisfactory, but there are still a few details that need to be changed. AI During communication, the exact location cannot be clearly stated.AI They couldn't pinpoint the target for modification, and after several rounds of discussions, the number of errors they corrected actually increased.
There's now a workaround for this problem. On GitHub... Cowart The project simply ported the infinite canvas to Codex. Users can draw arrows and write comments on the original image, which can then be edited by... Codex Generate a revised version.
Follow us on WeChat and reply with "open source",join inAIopen sourceProject discussion group
Cowart is a native infinite canvas plugin for Codex, based on tldraw.
Project address: https:////github.com/zhongerxin/cowart#
The canvas can store text, sketches, reference images, and generated images, and also allows you to directly annotate images with modification suggestions.
Canvas data is stored in the canvas directory of the current project. Codex can read the canvas selection and place the generated result in the specified location, saving the process of manually saving, uploading, and reformatting.
Cowart has a built-in... MCP The service currently offers two tools:
get_cowart_selection Used to read the currently selected canvas element, including the element ID and type, coordinates and rotation angle, width and height information, image file name and resource address, and whether it belongs to [the selected element]. AI Image placeholder.
insert_cowart_image The MCP service is responsible for inserting local images into the canvas. It copies the image file, creates a tldraw image record, and finds the appropriate position based on the anchor point.
Illustration orientation supports left, right, and bottom. When the target area is occupied by other content, Cowart will continue to search for empty space to reduce image overlap.
Cowart is suitable for article illustrations, WeChat official account cover sketches, social media materials, PPT visual schemes, product concept diagrams, and storyboard designs, all of which require frequent comparison and modification.
To install Cowart, simply enter the following in your Codex:Prompt wordsThat's fine:
Prompt words:
Please install the Cowart Codex plugin from https://github.com/zhongerxin/cowart.git.
Please clone the repository to ~/plugins/cowart and confirm that .codex-plugin/plugin.json exists.
To add a plugin to the personal marketplace, first run `codex plugin marketplace add ~`.
Then run the codex plugin add cowart@personal.
After installation, please verify the plugin and let me know if you need to open a new conversation to load new skills and MCP tools.
Prompt words：Open the Cowart canvas for this project.
Cowart will then start a local service, with the default address being:
http://127.0.0.1:43217/
Case 1: Generating an image in the canvas
The Cowart toolbar has a "AI The "Image" tool has the shortcut key A.
After clicking, the canvas will create an image placeholder with a default size of 320 × 220. The placeholder can be freely resized, and Codex will read the aspect ratio and generate an image according to the same ratio.
Prompt words:according toPrompt wordsGenerate an image in the Cowart canvas: Generate a storybook cover.
Once generated, the image will be placed in the placeholder frame and will move along with it. You can continue to generate or replace new versions in the same location while keeping the placeholder frame in place.
Codex can generate an image even when no placeholder is selected in the canvas. Cowart will find a blank space nearby and insert the new image based on the current page and the selected element.
Case 2: Editing images using arrows and text
Cowart also provides a "comment" tool, with the shortcut C.
Dragging the mouse next to the image will generate a hand-drawn arrow. Releasing the mouse allows you to directly input your modification requests.
After completing the annotations, capture a screenshot of the canvas containing both the original image and the annotations, and then send the screenshot to Codex. Codex will recognize the location of the arrows and the annotation text, generating a clean new image.
Let's try modifying the image generated above:
Prompt wordsUsing my Cowart annotation screenshot, a clean revised image is generated next to the original image.
The new image is placed to the right of the original image by default. The original image, arrows, and annotations are all preserved for easy comparison of each revision.
Of course, we can also modify multiple places at once.
Prompt wordsUsing my Cowart annotation screenshot, a clean revised image is generated next to the original image.
As you can see, the annotation for "xiaolongbao" (a type of steamed bun) seems to have been incorrectly edited because the current annotation modification relies on screenshot recognition. Blurred arrow positions, low screenshot resolution, or conflicting annotations can all affect the generated results.
AI Taking unedited photos is no longer a niche activity.
Grand View Research estimates that globally AI The image generation market was worth $349.6 million in 2023 and is projected to grow to approximately $1.08 billion by 2030, with a CAGR of 17.7% from 2024 to 2030.
Current raw photos AI There are indeed many suggestions, but it's still difficult to articulate the changes clearly, and it's also difficult to iterate on the image versions.
Cowart's value lies in this: the original image, reference image, annotations, and revisions all reside on the same canvas. Codex can read the selection and place the new image next to the original. There's no need to switch back and forth between folders, chat windows, and the editor; changes can be pinned to the specific location indicated by the arrow.
Cowart is more like a AI The image workbench presents what needs to be changed in a visible way, rather than a string of easily misunderstood text descriptions.
AI The process of generating images is fast enough; the next important step is to integrate the generation, selection, annotation, revision, and comparison processes into the actual workflow.
Original link:GitHub garners 2,000 stars, packing an infinite canvas into Codex