Text to Vector
We have introduced the revamped Text to Vector feature.
We have introduced a new model and would want to hear your feedback on the quality of outputs. Additionally, we have made changes to the user experience of the feature and worked on providing greater control to influence the generation of vector graphics. A few highlights:
The new modal interaction allows the provision of appropriate user controls to influence the generation along with the prompts.
The properties panel is offered more contextually, and the options related to Generate Vectors come up in the panel, once a generated object is selected.
Generated patterns are now housed under the Fill pop-up and there is a new tab for patterns.
-
Frank commented
Very improved, but I can't figure out how I can generate not frontal vievied subjects but rotated at an angle subjects. Thanks
-
TT Kiker commented
It's kind of fun to play with, but it is certainly not helpful in my day-to-day at this point.
-
Junior Rocha commented
The new model have generations with a better quality than old one.
Refence image is very nice to have some control over the generations.Sometimes when the generated image have empty areas inside, if those areas are the white.
I can use arrow keys on contextual bar to navigate with the cursor on the text, the text box collapse.Pattern Generation have too much distorced results, simple prompt with hearts and dog foot print gives a very distorced results, need some lucky to have good generations on that.
-
Alan Gilbertson commented
In particular, I found the reference image feature so completely unpredictable that I gave up. Of thirty or so iterations, only one came close to matching the reference.
There is also an apparent inability to draw a complete human figure without cutting off the feet and/or top of head which is quite baffling.
The large regression since the original iteration of text-to-vector is a surprise, because genAI images from Firefly have been steadily improving in quality, and this goes against the trend.
-
Derek Watson commented
Extremities such as hands when generating subjects are 'random'.
The background is improving, especially when used to extend.
Rating output and refining rather than regeneration might get you two intended result faster?
-
Rene Andritsch commented
Reading the help files about text to vector I would have expected to find the pattern functionality also in the Contextual Task Bar. I think users would benefit from also having access to the pattern creation there.
Also you could adopt the icons for scene, subject and icon as you already implemented this for the pattern in fill and stroke drop down.