Cool idea but if it is tied to the model vision quality, idk.. maybe when all models vision are like fable 5 onwards and gpt 5.6/6 vision, but until then this is strictly a niche use case, you save a lot of tokens by simply not using Fable for everything, and i understand the appeal tho, but this project depends on the model vision.. maybe you should invest on a way or a "companion" for pxpipe to ground the facts and reduce hallucination risk, cool idea, i just dont know if its reliable, i prefer to keep quality than reduce costs at certain point, depends on the trade off of course, but i dont see it being worth it here.
Cool idea but if it is tied to the model vision quality, idk.. maybe when all models vision are like fable 5 onwards and gpt 5.6/6 vision, but until then this is strictly a niche use case, you save a lot of tokens by simply not using Fable for everything, and i understand the appeal tho, but this project depends on the model vision.. maybe you should invest on a way or a "companion" for pxpipe to ground the facts and reduce hallucination risk, cool idea, i just dont know if its reliable, i prefer to keep quality than reduce costs at certain point, depends on the trade off of course, but i dont see it being worth it here.