Last week I hosted a coffee hour in the morning European time. Since many of you are asleep at that time, I’m planning to host another coffee hour later in the day. I’d like to know if you’d be able to join if I schedule it around 9 p.m. Central European Time. We can talk about Weavy, general AI topics, or anything else on your mind—it’s a relaxed, open vibe. I’d really appreciate it if you can have your cameras on during the session; it’s all about getting to know each other.

The aim here is to stress-test weavy to extract a single image from a storyboard whilst keeping all details exactly the same as they appear in the the smaller grid images. I have played around with using image upscales such as Topaz with limited success. The AI seems to hallucinate and fill in details such as decals on the bike and also the bike itself. After many conversations with LLMs- in case Claude it appears that the best solution is to independently create 9 single high res images. Weavy does not like trying to extract 1 single image from a grid in one sweep- it gets confused and the details are muddled at best. I'm putting this in the community to see if anyone else has come across this issue and their workflows to resolve this.

I always show clean workflows but that is not the reality in my projects :) Especially in the first phase of concepting and iterating. But I do make sure to do a little but of householding after a few days. Are your team clean or team messy?

People from Oregon to Serbia, 2AM timezones, cameras on, real conversations. That felt good. I walked through the rim workflow live and got some great questions along the way. Here are a few things that came up that I think are worth sharing: Always label your nodes and sources. Once you start copy-pasting images into a workflow you lose track fast. Sticky notes are your friend. Name everything. Test cheap, ship quality. I ran the whole workflow on a lower-res model first. Faster, cheaper, same logic. Once the workflow is locked, you upgrade the model. Not before. Routers keep your canvas readable. One input, multiple destinations. It sounds basic but it changes how clean your workflow looks. And a clean workflow is a workflow you actually understand three weeks later. Upscaling is harder than it looks. If you upscale a small panel image without feeding the original product references back in, the model will guess. And it will guess wrong. Feed it context, always. The workflow is the product. One of the attendees immediately saw a use case for second-hand car dealers in Serbia. That's exactly the point. You build it once, it works for a client forever. The recording is attached. Fair warning: I did not record it fullscreen. (I know.) So I will re-record the workflow properly and add it to the classroom later. Full screen this time, I promise. Thanks for showing up. Let me know what you think in the comments! Workflow: https://app.weavy.ai/flow/QIOqmzgXl2hV09GD54VWA3