Batch vision checks after generation, reject inconclusive answers, recommend llava
This commit is contained in:
1 parent
d9fc18fcca
commit
47433dbfa8
2 files changed
+64
-41
No files matched your search
+1
-1
@@ -153,7 +153,7 @@ const FIELDS=[
|
||||
["","Text and shared options","header","",""],
|
||||
["ollama_url","Ollama URL","text","","g"],
|
||||
["ollama_model","Ollama model","model:text","Writes scripts and tags. Models not installed yet are downloaded automatically","g"],
|
||||
["vision_model","Ollama vision model","model:vision","Checks generated scenes on made-for-kids profiles. moondream suits 16 GB machines","g"],
|
||||
["vision_model","Ollama vision model","model:vision","Checks every scene on made-for-kids profiles. llava:7b is recommended, moondream is too weak to judge anatomy","g"],
|
||||
["ollama_timeout","Ollama timeout (seconds)","select:120|300|600|900","Max wait per Ollama request before retrying","g"],
|
||||
["aspect_ratio","Aspect ratio","select:9:16|16:9","","g"],
|
||||
["target_seconds","Target length (seconds)","select:30|45|60|90|120|180","Split into scenes of the length below","g"],
|
||||
|
||||
Reference in new issue
Block a user