Want more intelligent visions of your inbox? Subscribe to our weekly newsletters to get what is concerned only for institutions AI, data and security leaders. Subscribe now
Adobe Photoshop is among the most famous programs that have been created ever, and is used by more than 90 % of creative professionals in the world, according to Optical.
So the fact that a A new open source form artificial intelligence – QWEN-DISE EDITWhich was released yesterday by the Chinese giant QWEN team for e-commerce, is an artificial intelligence researcher It is now capable of achieving a large number of functions similar to Photoshop with text inputs aloneIt is a noticeable achievement.
The teacher built 20 billion The QWEN-TAIGE Foundation Model was released earlier this monthQWEN-TAINGE-Edit extends the unique strengths of the system in providing the text to cover a wide range of editing tasks, from the micro-appearance changes to the broader semantic transformations.
Simply download the start image – I tried one of myself The last annual conversion conference for Venturebeat In San Francisco-then write instructions for what you want to change, and QWEN-remitit will return a new image by applying these modifications.
Artificial intelligence limits its limits
Power caps, high costs of the symbol, and inference delay are reshaped. Join our exclusive salon to discover how the big difference:
- Transforming energy into a strategic advantage
- Teaching effective reasoning for real productivity gains
- Opening the return on competitive investment with sustainable artificial intelligence systems
Securing your place to stay in the foreground: https://bit.ly/4mwngngo
An example of an input image:
Example the image of the directing with a wave: “Make the man wearing an evening.”
The model is now available on several platforms, including QWEN Chatand Embroideryand Modelsand JaytabAnd through Alibaba (API) programming interface (API)The latter, which allows any developer or third party institution to integrate this new model into its applications and its workflow.
I have created my examples above QWEN ChatThe QWEN team competitor on Openai, however, it is worth noting which ambitious users are limited to about 8 free jobs (input/outputs) in a period of 12 hours before reseting them. Users can pay access to more jobs.
With the support of both English and Chinese inputs, and the dual focus on both semantic meaning and visual loyalty, QWEN-Center-Edit aims to reduce barriers to creating visual content content.
Given that the model is available as an open source symbol Under Apache 2.0 licenseIt is safe for institutions to take, download and prepare them for free on their own devices or clouds/virtual machines, which may lead to great cost savings from royal programs such as Photoshop.
like Junyang Lin, a QWEN team researcher on X, wrote, “He can remove a strand of hair and modify very sensitive images.”
The team’s advertisement reflects this feeling, as QWEN-Went-Edit offers not as a completely new system, but as a natural extension of the QWEN image that applies its unique text and the double coding approach directly to the editing tasks.
Double symbols allow edits to preserve the pattern and content of the original image
QWEN-Dise-Edit builds on the basis of its creation QWEN-DestIt was presented earlier this year as a large -scale model that specializes in both images generation and text presentation.
The high technical report of QWEN-TAIGE has its ability to deal with complex tasks such as displaying texts at the level of paragraph, Chinese and English letters, and multiple lines with accurately.
The report also confirmed a Double coding mechanismSimilarly feeding images in QWEN2.5-VL for semantic control and variable automatic encryption (VAE) for restorative details. This approach provides adjustments that are still sincere for both the intention and the appearance of the original image.
These architectural options themselves support QWEN-IMAGE-Edit. By taking advantage of the dual codes, the form can be adjusted to two levels: Semantic amendments Who changes the meaning or structure of the scene, and Amendments to appearance Which provides or remove the elements while maintaining the rest without touching.
Semantic It includes the creation of new intellectual ownership, 90 or 180 -degree rotating objects to detect different views, or convert inputs into another style such as the GHibli studio. These adjustments usually adjust many pixels but they maintain the basic identity of the organisms.
here An example of semantic liberation From Shridhar Athinarayanan, an engineer in AI App platform, who used a host or “deduction” application for QWEN to shoot a picture of Manhattan to look like a LEGO game.
Editing appearance It focuses on accurate local changes. In these cases, most of the image remains unchanged while changing specific objects. The demonstrations include a remarkable banner that generates a reflection in the water, removes the loose hair strands from an image, and changes the color of one letter in a text.
One of the good examples of the editing of appearance with the editing of QWEN-TAIF comes from the co-founder and CEO of Assywai Thomas Hill that published a Along with x His wife appears in her wedding dress below a corridor and another with the same corridor covered with graffiti:
Besides the firm power of QWEN in presenting the Chinese and English text, the system that focuses on liberation is placed as a flexible tool for creators who need more than simple obstetric images.
Double control of the semantic range and sincerity of appearance means that the same tool can serve completely different needs, from the creative IP development to the re -texture of images at the production level.
Add or remove the text to the pictures
Another prominent ability is Editing text bi -language. QWEN-WENENT-EDIT allows users to add the text, remove or modify the text in Chinese and English while maintaining the line, size and elegance.
This expands the reputation of the QWEN-IMAGE image to present a strong text, especially in difficult scenarios such as complex Chinese characters.
In practice, this allows accurate editing of posters, marks, shirts, or artworks of the calligraphy where the small text details are concerned, as shown in Another example of the symmetric copies below.
One of the error correction demonstration included a part of the Chinese line that was created through a step -by -step editing process.
Users can highlight incorrect areas, direct the system to fix them, then improve the details until the correct letters are provided. This repetitive approach explains how the model can be applied to the tasks of release high risk as the accuracy is necessary.
Applications and cases of use
The QWEN team is the most prominent set of possible applications:
- Creative design and expansion IPSuch as generating the mascotal symbols based.
- Advertising and creating contentWhere slogans, banners and heavy textual visuals can be customized.
- Virtual and art AvatarWith the transmission of the style, support unique personal representations.
- Photography and personal useIncluding background adjustments, clothing changes, and removing the object.
- Cultural memorizationIt appears by correcting the classic font works.
By bridging the exact liberalization with the broader creative transformations, QWEN-WINENT-EDIT meets professionals who need control while staying friendly to informal experimentation.
Measurement and performance
According to the QWEN team, the assessments via general standards indicate that QWEN-IMAGE-EDIT is providing Later In photo editing.
This is followed by the broader technical assessments in QWEN-HIE, where the basic model has achieved leadership results in each of the tasks of generating public images and the tasks of presenting the text.
While the editing numbers specified in the version have not been detailed, the QWEN image itself is largely classified as independent assessments such as Ai Arena, as human residents have compared the outputs through models from various service providers.
API pricing and its availability
during Alibaba Cloud Studio StudioDevelopers can access QWEN-Image-Edit as an application programming interface. Pricing was set in 0.045 dollars per imageWith a free share of 100 pictures valid for 180 days After activation.
The service is available at the beginning in Singapore areaWith a rate limit Five requests per second Even Two simultaneous tasks for each account.
To use API, developers must get the API Studio Model and can call the model via HTTP or through SDK Dashscope in Python or Java.
Pictures can be presented as URL or Base64 format, with supported decisions ranging from 512 to 4,096 pixels and file sizes of up to 10 MB. The output images are hosted on the storage of the Cloud Alibaba object with 24 -hour valid links, which requires users to download and save the results immediately.
What is the following for QWEN?
QWEN puts the picture as a stepD Reducing barriers to create visual content. By making careful and harmonious editing easier, the model Applications can be supported from design studios to informal users who improve personal projects.
The system also indicates a broader trend in developing artificial intelligence: bypassing tools for individual purposes towards tools that merge liberation, correction and improvement.
With both semantic flexibility and accuracy at the level of appearance, QWEN-WENENT-EDIT reflects this shift, mixing the gynecological strengths of large models with the reliability needed for professional editing.
Source link