News for hackers. Stuff that matters.
Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @07:11AM (#49386414)
Oof 800 by 800 kills a lot of use cases
Re:DeepSeek-v4-flash-vision-exp by johndough on Friday August 21, 2026 @07:33AM (#49386565)
Might still be fine. The most recent crop of vLLMs proactively use whichever programs are available on the system (e.g. ImageMagick or PIL) to "zoom in" by cropping subimages if they can't quite make out the details.
Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @10:03AM (#49388233)
Downsizing a higher res image to lower res means the zoom will be blurry.
Re:DeepSeek-v4-flash-vision-exp by andai on Friday August 21, 2026 @10:19AM (#49388513)
They process the original image file with Python on the local device. (And I've seen the web chats do this with their "computer use" features too.)

The really wild one is even blind models will do this and they'll try to run stats on the pixels to figure out what it looks like... the even wilder thing is that it kind of works!

Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @11:24AM (#49389544)
If the API accepts only 800 by 800, the aegument youre making is "fix it in the harness".

I don't think the n by n subgrid fixes this the way most harnesses do, as it'll fail to count things if you have more overlap and fail relatiomships if you have less

Re:DeepSeek-v4-flash-vision-exp by tjoff on Friday August 21, 2026 @12:55PM (#49390834)
That seems weirdly specific?

And if you are counting things it should be trivial to note the position of your items and not double-count them, no?

Re:DeepSeek-v4-flash-vision-exp by adastra22 on Friday August 21, 2026 @11:56AM (#49390019)
They’re not talking about zooming, hence the quotes.
Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @12:27PM (#49390460)
If the harness does it that's just like saying "please use a workaround". You'll lose fidelity and LLMs will lose the ability to count things or maintain relationships for schematics, etc
Re:DeepSeek-v4-flash-vision-exp by johndough on Friday August 21, 2026 @01:05PM (#49390992)
LLMs read images by splitting them up into e.g. 16x16 patches, which are then converted to embedding vectors and fed to the LLM, so from a technical point of view, feeding a big image as many 20x20 patches all at once is not too different from cropping subimages from the image, splitting those subimages into patches and feeding them to the LLM. Of course, the LLM has to be trained to understand that those images belong together, but it can be done.
Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @03:15PM (#49392639)
You say "it can be trained" but they fail at counting in my use cases, let alone maintaining symbolic relationships.

Can you point me at one that can understand a detailed block diagram? Frontier is fine, soliciting recommendations

Re:DeepSeek-v4-flash-vision-exp by johndough on Friday August 21, 2026 @06:10PM (#49394322)
For counting, there are specialized counting models, e.g. https://huggingface.co/spaces/MengqiLei/count-anything-demo

I tried to parse hand-drawn ER diagrams in the past and did not have much success with any model, frontier or otherwise. If you have annotated data, I'd recommend finetuning a recent (dense) VLM, but don't expect 100% accuracy. https://unsloth.ai/docs/basics/vision-fine-tuning

Re:DeepSeek-v4-flash-vision-exp by knollimar on Saturday August 22, 2026 @12:28PM (#49401287)
good advice; tbh I'm really trying to use counting as a proxy for "understand symbol, flag annotation next to it, associate annotation with symbol and store as object in location".

Ideally, look at second diagram and see something similar in the same location and understand it's the same physical object.

Re:DeepSeek-v4-flash-vision-exp by johndough on Friday August 21, 2026 @12:56PM (#49390851)
Yes. When the LLM tries to read an image, it will be resized by DeepSeek's server to 800x800, which might be a bit blurry. The LLM will then crop a smaller image from the high resolution image (using e.g. the `convert` tool via bash) and will then read the small cropped image. This image will still be resized to 800x800 by DeepSeek's server, but since it is already small, there is no or little loss of quality.
Re:DeepSeek-v4-flash-vision-exp by johndough on Friday August 21, 2026 @01:24PM (#49391257)
The order is:

    LLM issues tool call to read high res image ->
    harness sends high res image to server ->
    server downsizes it to 800x800 (blurry) ->
    LLM issues bash command (e.g. `convert`) to crop a small subimage (e.g. 600x600) from the high res image ->
    LLM issues tool call to read subimage ->
    harness sends subimage to server ->
    server does not resize the subimage because it is small already, so it is not blurry when finally ingested by the LLM
Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @03:13PM (#49392627)
Then you have a separate issue where the LLM can't piece together 9 subimages well.
Re:DeepSeek-v4-flash-vision-exp by nwienert on Friday August 21, 2026 @10:36PM (#49396084)
Insert the famous Louie CK phones on planes bit. Sand is thinking, working, coding and now looking for you, for pennies an hour... but "oof" it's not good enough.
Re:DeepSeek-v4-flash-vision-exp by wongarsu on Friday August 21, 2026 @07:31AM (#49386554)
For most use cases you can fix that in the harness. Just give the model a tool to request a crop of specific coordinates of any image it has in its context. Call the tool "zoom" and it should be intuitive for the model

Maybe there are some use cases where you need high detail everywhere at once, but for OCR of small text and the like a zoom ability should be sufficient

Re:DeepSeek-v4-flash-vision-exp by embedding-shape on Friday August 21, 2026 @07:57AM (#49386773)
For really dumb models I've also had success automatically cropping it into a grid of N images with the max size, then processing each cell individually, then once all been processed, do one final call with resized image + all other context previously generated per cell. Basically a workaround to the image dimension restrictions without loosing fidelity. Works well with even dumb 7B models.

Can't remember if I stole this idea from some existing public harness though, can't remember. If someone knows of public harnesses that do this already, please share them :)

Re:DeepSeek-v4-flash-vision-exp by dotancohen on Friday August 21, 2026 @08:40AM (#49387222)
Does this not loose context? Especially e.g. in fonts where the character pairs 0O 1I 1l Il may be difficult to differentiate?
Re:DeepSeek-v4-flash-vision-exp by skeledrew on Friday August 21, 2026 @08:57AM (#49387420)
That's what the grid crop should handle. The detail is retained at that level, and then everything is logically stitched together again using the lower-res-full-image as reference. That's going to be 2x token usage at minimum though.
Re:DeepSeek-v4-flash-vision-exp by Aeroi on Friday August 21, 2026 @03:23PM (#49392742)
i think the claude cookbook has a file that does this, called tiling.
Re:DeepSeek-v4-flash-vision-exp by shadyr on Friday August 21, 2026 @07:40AM (#49386630)
It might also be due to its experimental status. Wouldn't surprise me if the GA version allows for larger input. Either that or the eventual pro version.
Re:DeepSeek-v4-flash-vision-exp by asdfsa32 on Friday August 21, 2026 @07:18AM (#49386460)
flash vs fine details. Pick one.
Re:DeepSeek-v4-flash-vision-exp by Doohickey-d on Friday August 21, 2026 @07:32AM (#49386557)
Gemini "flash" models have an option for media resolution, including a high resolution option for screenshots.
Re:DeepSeek-v4-flash-vision-exp by skeledrew on Friday August 21, 2026 @08:53AM (#49387361)
At what price point?
Re:DeepSeek-v4-flash-vision-exp by Chnmy on Friday August 21, 2026 @09:50AM (#49388045)
what are these use cases?
Re:DeepSeek-v4-flash-vision-exp by knollimar on Friday August 21, 2026 @10:04AM (#49388245)
Anything where there are symbols representing in space (e.g. schematics). Thats pretty broad
Re:DeepSeek-v4-flash-vision-exp by stronglikedan on Friday August 21, 2026 @12:35PM (#49390574)
I don't know about a lot. Probably more like a few. I take a lot of screenshots for various reasons, and over 800 seems like I could have done a better job framing and cropping.
Re:DeepSeek-v4-flash-vision-exp by hedgehog on Friday August 21, 2026 @04:54PM (#49393664)
Multi-turn, expose a tool that extracts crops. Works great.