iiinova — where your digital self keeps evolving

It appears to be working. 🤖👍

000

Using these systems is kinda funny. Copilot with free tier Claude didn't code it properly. Seeing of Gemini 3.1 Pro can fix it. 🤣👏

010

Got it running again. Copilot was writing script commands into the solution file. 😅

000

VS2026 and Copilot totally borked the project, somehow destroying the solution file (and potentially other files) ope 😅

000

Scott @twoc.bsky.social · 04/07/2026

Having multiple AI work on getting a console log control to stop scrolling when the user scrolled up is apparently an extremely difficult task. 😅😄🤷 used up the monthly Kiro credits and now having Copilot try to finish the task. 🤖🤖

400

I wanted to move embedded YouTube videos in website articles down to the bottom of each article. Another easy project by using Gemini Pro to make it. A few checks to verify and had it process all of the websites. I'd eventually like to reprocess affiliate links to fix old ones. 🤓

000

I'm by no means a domain expert in this stuff, so my "job" is trying to give it the knowledge I'm able to collect and simple logical guidance. I can also research things in the moment, but yeah. 🤓🤖

000

That compute didn't last long. Also, it isn't very smart. It failed at multiple attempts to fix something it made.

000

This is pretty good so far. We'll see how long the free tier compute holds out.

000

Scott @twoc.bsky.social · 24/05/2026

I'm trying yet another agentic programming IDE. 🤖😄 Every free tier I can find to get the job done without having to resort to hand-coding. 👏

200

I'm working toward trying to make an ACDSee Photo Studio plugin that does automatic white balance and exposure (highlight & shadows adjustments also welcome). I made two XMP files with the range of sliders in the Develop tab so I can avoid data issues.

000

Scott @twoc.bsky.social · 22/05/2026

Another quick stream focused on trying to get ACDSee to register RAW image changes. It looks like it is working! That's one big piece of functionality. Now I can focus on getting white balance and exposure adjustments actually looking good. 🤓😽 VOD: youtube.com/live/w7OmQOq...

youtube.com

Twitch news? Researching the ACDSee metadata insertion issue & trying to get it working. 😎🤓😺

YouTube video by Scott 🖥️ The Way Of Coding 🖥️

000

Scott @twoc.bsky.social · 21/05/2026

Oddly, adjustments are applying to a folder of 5 images, but not of one that's got more pictures in it. 😅 Gonna say it again, programming is my souls-like game, but's spending a game's worth of time on it is normal (30-70 hours most of the evening I was working on it and it's now tomorrow 👏).

000

Scott @twoc.bsky.social · 21/05/2026

Past 3am, getting correct XMP/sidecar/develop settings has been difficult. Even the most advanced AI I have access to needed direction. I've got it applying changes now, but it needs a lot of work. Especially with white balance calculation. Also, there are still problems applying changes.

100

Scott @twoc.bsky.social · 18/05/2026

Turns out my use of llama.cpp had issues I wasn't paying attention to with the command. It looks like tokens per second jumped from like 15/s to 100/sec.

000

Scott @twoc.bsky.social · 18/05/2026

I've got a bunch of convention photos to edit, but it's all practical RAW image edits. ~7 months ago I did some initial research on Acdsee plug-ins. Right now I starting the initial project, but ran out of credits on the cloud models. Seeing if my local option can finish the work.

100

Scott @twoc.bsky.social · 15/05/2026

Another quick programming project (that Copilot AI wrote based on my design document and some refinement/fixing up afterward). Finding projects in my project drive that are finished based on if there are horizontally formatted mp4s in a Renders folder. 🤓🤖

000

Wow, it is actually working as far as I can tell. The task I gave it was pretty simple but it figured it out and wrote the code properly! 🤖😎 It wasn't fast but it also wasn't crazy slow, so maybe I might be able to use it in a pinch. But I still want to try with smaller models.

000

This might be working now. Gemini Pro created a go-between of sorts to take the breaking messages from VS Copilot and feed it to the local model in ways that it can handle. This prompt ran at least! It's slow on my PC so not especially useful, but I'll try it with a smaller model out of curiosity.

100

I used Copilot (Claude 4.5 mostly & a bit of GPT) to make it. Now I'm using Kiro to fix up the comments that Copilot didn't bother adding. I have it set to auto mode, so not sure which model that one is using. Ideally this level of usefulness would be viable locally at some point.

000

A few minutes writing up a design document to make an application that transfers video renders from where I put them back into their respective project folders. This should save a lot of time because I had done it by hand. I had 80 or so renders sitting waiting to be transferred, EZ! 🤖😎

100

kind of works but too slow and probably not smart enough. It's been minutes of "Look over the code and see if you can find any functions that are missing a /// type xml comment at the top that describes what a given function does...."

000

Now it's trying to make its own server to go between the two instead of using LiteLLM because that wasn't following instructions. 😅🤖

000

Another branching path, using LiteLLM as a go-between. We'll see if it works!

000

speaking of failure responses... ope.

000

A ton of changes and trying different commands Google AI suggested later, it's giving me a response. I also switched to the MoE Qwen 3.6 instead of the standard one. The lack of VRAM is rough and this one is faster at least. Trying to push it to see if it sends a failure response to VS Copilot.

400

Scott @twoc.bsky.social · 03/05/2026

Trying out Qwen,3.6 27B 4-bit on my hardware. It's slow but it runs. If I had a GPU with 32GB of VRAM it might be usable. The AMD Radeon AI PRO R9700 is around $1400 and the Intel B70 is $1100 but not sure on support of either.

100

Maybe I deleted the post, not sure what I said. If it's Gemini Pro in AI Studio, I'm the go-between which is safer. But trying Kino, it can be fully automated w/ more risk. That said, there is a lot of risk and I think you should do heavy test driven development & review if you attempt that route.

000

Scott @twoc.bsky.social · 03/05/2026

Yeah, I definitely can't afford thousands of dollars to get enough VRAM and compute locally, but I would prefer that solution long term. I'm using free online solutions at the moment. I'd hope into the future local is the default, but concerned it will be the opposite.

Scott @twoc.bsky.social · 03/05/2026

I'm impressed with the state of AI assisted programming. It's come a long way since trying to use the original ChatGPT to this Kino program basically turning into and advanced worker I direct and help guide toward solutions it usually finds and implements.

000

Open source AI exists, but the current climate has made it extremely expensive and difficult to do locally. Sure you can code by hand, but IMO that goes against the whole idea software developers that use computers to facilitate their goals. Otherwise why re-use functions, etc.

This and Copilot free tier are showing issues with software development into the future. It's a huge shift toward making devs play to play. Big tech is hogging hardware (compute) which inflated prices of key components. Then turn around and try ration/sell it.

I've used 173/500 of the starter credits so far. The free tier only gives 50 per month, so in the future I won't be able to do a ton with this. ~173 was adding/fixing comments, design, first fixes, and re-work.

100

For that design back-and-forth with design and code analysis it was great. For coding there was a list of needed fixes, it failed at some of it which required going into the changes and having it re-work them. Probably more of an issue for the "auto" model selection than the IDE itself.

100

It can't replace Visual Studio, but I'd say it's more functional feeling than the Copilot panel in VS 2026. I feel like Gemini Pro in AI Studio is a bit better at coding, but the agentic approach feels convenient. I spent hours using it to re-design and structure future work plans.

200

Been trying out Kiro. A friendly chatter stopped by a stream recently and suggested it. They apparently work at Amazon, so not surprised they use it. Good dev talk & they even gave some bits which probably the first time I've gotten any. 😺 So far Kiro is pretty cool as a supplemental IDE.

100

Had a breakthrough getting dock-able "Custom Command" scripts working with Vegas Pro. 🤓

000

🤔 This Nemotron Cascade 2:30b seems to be running okay on my PC with its meager RTX 4060 8GB GPU. The CPU is an AMD 9900x and 64GB of VRAM. Slow part is the initial start.

000

Hmm, the AMD Radeon AI PRO R9700 looks like the most reasonable option as a single GPU. The B70 is cheaper, but it doesn't look like the software support is there? Though, I don't know how AMD compares to Nvidia for this use case. www.amd.com/en/products/...

amd.com

010

Scott @twoc.bsky.social · 17/04/2026

I'm trying to implement a dockable window in Vegas Pro, but there are a lot of technical challenges. Though I'm not sure if it's Gemini Pro failing at understanding how to edit video on the timeline or some other issue/bug. Giving Copilot with Claude a chance to see what it can do.

010

My transcription tab uses an open source offshoot of OpenAI's Whisper API, but gives word level timecode accuracy. That makes it more viable to use for automatic video editing. github.com/linto-ai/whi...

github.com

GitHub - linto-ai/whisper-timestamped: Multilingual Automatic Speech Recognition with word-level timestamps and confidence

Multilingual Automatic Speech Recognition with word-level timestamps and confidence - linto-ai/whisper-timestamped

000

Scott @twoc.bsky.social · 17/04/2026

I'm working on a big project that is a successor to my Vegas Pro video editor scripts. In this case I'm taking things a step further to manage video projects from start to finish. Hopefully with a lot of features that facilitate producing work faster while keeping it high quality.