[Redefine] Image Description Issues & Improvements

I’ve been using the redefine feature now quite heavily for past few months and pretty impressed but it seems to me recent version 8.3.3 seems to have put far far less weight on the “Description” field even when creative is 3,4 or higher. I used to use 3 or 4, because 5 would tend to change the color and “sharpen” stuff way too much, so 3 and 4 where the happy setting.

So at 3 or 4, I could easily crop into a face, write “blue eyes” it would actually 90% of time do a great job of changing eye color.

That doesn’t happen now until you force it to 5 or 6 at which point image is too over processed and starts to take on a green hue to everything to look good (you guys might want to dampen that effect down a touch). Not sure why increase "creativity would change entire image more and more to greenish ugliness.

The other thing I’ve seen is there is a very heavy “sharpen” tendency now where there wasn’t before even with sharpen at 0. Before you can add detail but it was a nice photographic or smooth effect to it. Now it’s just tends to create really harsh jagged edges to bodies, shoulders, hands fingers.

So whatever took place to make some changes… what you had before was better :slight_smile:

And the Color shift at higher creativity, that’s been there from the beginning. No color change at 1 and gradually images take on a greenish tint or reduction of red as you increase creativity. That’s super annoying because means you are forced to color match back in PS to what might be the original colors.

1 Like

1. Scrolling Selection Issue in the Prompt Field

  • Description (Revised): When I try to click-and-drag to highlight text inside the “Image Description/Prompt” box—and the prompt is longer than the visible area—the text area scrolls as I drag. However, the highlighted selection stops following the line of text once it scrolls out of view. In other words, the cursor no longer tracks the moving text, so I can’t highlight and delete content beyond the initially visible portion.
  • Steps to Reproduce:
    1. Open the Topaz desktop app (e.g., Video Enhance AI v3.8.1).
    2. Load or paste a prompt that spans multiple lines—longer than the height of the prompt box so a scrollbar appears.
    3. Click at the beginning of the prompt, then drag downward to select. As the text scrolls up, the visible caret loses alignment with the text under the cursor, preventing selection of lines further down.
  • Impact: I cannot reliably remove or edit sections of a long prompt. Since most of my prompts exceed the visible window, this makes it extremely cumbersome to refine wording, because I can’t highlight and delete text past the first screenful.

2. Lack of Local Queueing for Multiple Unique Prompts

  • Description: In Cloud Render mode, I can submit multiple images—each with its own distinct prompt—to a queue and let them process in the background. In contrast, with the local (desktop) renderer, I’m forced to wait for one image to finish before starting the next prompt-and-render job. There is no local queue that supports separate prompts per image.
  • Steps to Reproduce:
    1. Switch to Cloud Render mode and submit Image A with Prompt 1; then, while it’s rendering, add Image B with Prompt 2. Both images process independently.
    2. In the desktop app’s local rendering mode, submit Image A with Prompt 1. While it’s processing, attempt to start Image B with Prompt 2. You cannot—Topaz requires A to finish before B can begin.
  • Impact: I often need to batch-process dozens of images, each requiring a unique prompt. Without a local queue, I’m forced to babysit the app—waiting for each job to finish before I can kick off the next. This dramatically slows down my workflow when I lack reliable internet or cloud credits.

System & Environment:

  • Topaz Version: Video Enhance AI v3.8.1
  • Operating System: Windows 10 Pro (Version 22H2)
  • GPU & Driver: NVIDIA GeForce RTX 3080, Driver 531.18
  • CPU: Intel Core i9-10900K @ 3.70 GHz
  • RAM: 32 GB DDR4

Requested Fixes & Improvements:

  1. Prompt-Field Scrolling Selection
  • Ensure that when dragging to select text beyond the visible area of the prompt box, the caret stays locked to the line under the mouse as the text scrolls. This will allow continuous highlighting and deletion of text in long prompts.
  1. Local Render Queue
  • Add a “Batch Queue” feature in the local renderer so users can assign different prompts to multiple images in a queue.
  • Allow queued jobs to run sequentially or in parallel without blocking the UI, and enable reordering or cancelling jobs while previous ones are processing.

Please let me know if any additional logs, recordings, or information would help reproduce and resolve these issues. These fixes would significantly improve my ability to refine prompts and batch-process images efficiently. Thank you for your time and continued support.

Sincerely,

Brad (The Shaman / Mister Media)
the.schultz00@gmail.com

1 Like

Your post seems to have been auto sent to me - possibly because I have posted several times in other sub-topics on each one of the things you report here.

Other than offering sympathy I don’t think there is anything available to help at the moment.

My guess is that following the Jan 2025 announcement of 130 new hi-tech jobs at Topaz a lot of people may be running round in circles trying to find out where they are supposed to be sitting - never mind what they should be doing :smiling_face_with_sunglasses:

Sadly patience needed until normal service is resumed … :sleeping_face:

Hello!

Thanks for the detailed comments.
Taking the time to share the context & intent is a huge help and is a great example for others to imitate.

The developers are actively seeking improvements to the Image description parameter.

Until updates are released, there are some excellent tips in this thread:
[Redefine] Image Description Tips

Specifically on navigating the text field, I recommend

  • click once to place the cursor,
  • scroll to the desired point,
  • hold shift, then
  • click where you want your selection to end

This works in either direction (up or down) and is a temporary work around to the issue of drag-selection.

Personally, I keep a note open to write and edit my image descriptions, then copy and paste into the text field, avoiding all of these quirks.

Hope that helps for now.
Thanks :nerd_face:

1 Like

I confirm the feeling that the previous versions produced better results; moreover, even with only “Medium” creativity, it completely changes people’s faces.