I appreciate and hear your thinking here. But if I can gently push back: unless a person has a truly novel, brand-new idea, we all regurgitate what we have learned. For most people—not all—the best we can do is reformulate what we know or find a unique application for the knowledge we already have.
When it comes to its knowledge base, AI is essentially a fancy set of encyclopedias. Yet we don’t think twice about asking someone to “look it up.”
Change keeps on changing.
And I’ll push gently, too. Yes, many of us are teaching what we’ve learned and obviously that came from somewhere. However, it’s our unique voice that makes for truly new content. Just as I try to develop a style of my own for my playing, or for my photography, I also do so with my writing. The other aspect that tends to come from creative, rather than AI, writing or ideas is cross-pollination. While you could try to push a Large Language Model to do this (e.g. “write me an essay on X discipline but I want you to integrate ideas from Y discipline”), it’s the truly unexpected crossovers that tend to generate really new content. Most people using AI for writing are just asking for it to “write nice about…” under the assumption that the AI will come up with something more intelligent sounding than they would. I coach people to write things themselves, and then, and only then, have AI analyze it and make suggestions. This is a bit like the writer and editor relationship.
I like this. I think that is totally fair.
Most of my own writing these days is aided by AI. Research, editing and proofreading. That said, I am so particular about what I publish that my most recent project has been in the works first more than year. Much of it is in its 10th-12th revision. The initial draft took less than a week. That was over 1,000,000 words. It has been edited down to a bit over 375,000 words. Most of the extra was AI fluff and ugly prose.
So I appreciate what you teach.
Right now it feels like Christmas 
Unbelievable presents 
It’s like a whole different world… I used to be stymied by stock video footage and only being able to get those shots that were offered.
Here I can zoom - choose camera angle - sets - costumes and everything else related to what’s in my mind and have it realized
To me - Amazing - absolutely amazing
I see this as a way to final realize my vision of the music video
The drag is finding the right balance of video size and then topaz video upscaling…
I was at .9 that took three plus hours - but the content was great
Went down to .5 and increased to 10 seconds to 20+ minutes - but when upscaled in topaz was a little soft and too much decrease in quality
Now running my first ‘batch’ of three at .07 - I’m certain this will be the sweet spot
But this is taking a long time - so far….
But I can work on the spread sheet of camera shots and scenes while it’s rendering
Then 18 or so more shots to render - might be like the old days - start a video working and come back in the morning (yes used to do that in Final Cut express)
Thanx again Thom - really fun for me… I try to sleep but then ideas float in and I want to try them…
Haven’t touched the computer all day- it’s been creating - never know when it will finish - batches of 9 videos at a time
Taking 14 hours of rendering in comfy to make 36 clips for 3 minute music video - really seems to take longer as I never know when it’s finished rendering
At least 14 hours of pure processing time - probably 18 hours of no machine usage….
20 GPU cores is a little slim… I didn’t think I needed a lot of GPU cores for my audio work and website work
If I continue - I would upgrade to the studio M7 Max with 64gb RAM (or more) and at least a 2 TB SSD…
It is gratifying to see the rendered videos - amazing!
Waiting - for me - is hard - as is not using the machine for other tasks…
Last post - I learned I can create almost unlimited workflows and unlimited csv camera and scene prompts - then stack the run/work orders and walk away…
For the first video Chat GPT was guiding me step by step through the entire process…
Stacking 36 videos across five workflows is really nice 
Establishing shots, main shots, then in this case couple shots…
Sorry to post so much but excited about this new - to me - tech
I appreciate all the feedback here. My approach to using Claude Ai is research, editing and proofreading and then posting to the CMS. I use it as a sounding board and also as a think tank. Two things to understand is it typically will argue the majority position based upon data (although the majority position doesn’t mean that its the right one) and if you oppose it, it is predisposed to agree with you. So I spend a great deal of time arguing with it. You also have to watch out for the fact that it will sometimes assume from a small dataset, so if it uses quantifying language you have to force it to verify the claims.
So I’ve learned that you really do have to train it to get what you are looking for. I have a number of governing rules that keep it constrained to the way that I work and mostly think, although I do allow it to offer objections that others may raise that I haven’t thought of. It has taken me months working with it to refine the rules and it is an ongoing process.
The thing I appreciate most is it is more structured than I am. I always had trouble outlining in school and knowing when to start a new paragraph. If left to myself I’d write one long 300 page paragraph.
I stack prompts in AI all the time. The results are much better than one long prompt it has to figure out. My experience is the AI gets lost or simply doesn’t finish a task before moving to the next when it is too complicated. Claude seems to be best at multi-step prompts so that is my default. Sometimes if I am doing a series of articles I just have AI run through a checklist for each article.
This is absolutely true. I think because we get better at prompts. And because we get better at learning about how AI works (thinks). And because the models are trained and taught with more and more data points all the time. And because the AI models keep on changing, in theory getting better.
For me the hard part, because I love discovering new stuff, is to say good enough and use what I have. Agreeing to with myself to readdress it later if it truly needs to be better.
Just in case you missed it: Claude Code | LM Studio.
Your like magic man!!! Hahaha I did miss it…
I have both bionic studio and lm studio
Could not get bionic to stop thinking 
it would take hours to get a response
Spent an evening with chat gpt loading ejecting and trying different settings
I’ll post the settings and code inserts if anyone is interested…
Finally got it working at 10:30pm almost instantaneous - code: thinking was changed from xhigh to low - going to try ‘medium’ today - may get better results- my prompt is huge on cleaning the text - took an hour just to craft the prompt…
Update on video:
Redoing with stricter prompts and 15 seconds each
Clothes were different in each clip
Weather was different in each clip
Coffee cup - central to this video - was different in each clip
Coffee shop changed in each clip
Hair color changed in too many clips
Talking mouthing words - oops
Etc…
Too many issues…
Redoing 50 videos at 15 seconds each - with continuity and strict instructions for each video
This will take a long long time - I was wrong about the other video it was more along 20-24 hours (I thought it was done at 14 hours) this could take days to render…
Spent a good 4 hours on preproduction get images and prompts ready….
Started batch creation around 1:00 Chicago time August 31 - we shall see 
how long this takes…
But I’ll have consistent footage to make a coherent video this time - I hope 
Any links to the some of the videos?
Message me your email and I’ll send the first one that won’t be released….
Redoing… too disjunct…
Wasted three hours - columns on csv spreadsheet were wrong - rendered wrong material…
Still learning
Tightened prompts further and restarted again 4pm
5:07
Took 1 hour 7 minutes to do 15 seconds of 24 frames per second video clip
Times 50 clips
55+ hours of generation time…
Wow!!! A long time…
Again starting over 
was overly aggressive in the prompt asking to start in the scene… the image to Video translation never took hold and it’s not the same face or hair etc… starting over with new prompt
Well - wasn’t loading the new prompt - thought it was - deleted all old versions off the hard drive put a v2 on the new ones restarted the computer and comfy - and then the new prompts came up be an hour before I see what I got… hopefully right this time
Lots of test runs…
If it can’t see the face in the first frame - it makes it up
So it has to be sized correctly - took 3 hours to figure that out - the prompt has to be specific but not over hyped - a few sentences…
The images cannot be ‘tall’ and must fit in a 4K resolution
9:35 pm Chicago time fixed all - now running for 50+ hours - I hope
So we shall see what is there in the morning and what is there after two days…
Scrapped the run - still not rendering the face correctly
2) solutions - trying no transparent png and the exact video size of 1152x640 with the person inside and scaled back to 10 seconds instead of 15