Would anyone be so kind as to recommend models?

[Chimal1994]

Would anyone be so kind as to recommend models?

I've been using the ones in the workshop for a while (Gemma 2 and some image generation ones whose name I don't remember and it's not very relevant XD) and I also ended up trying Gemma 3, but it was disappointing. It ignored things and didn't track some variables. Gemma 2 turned out to be better, ironically.

In the workshop, I see stories with really nice images, and I'd like to know what models they use. Also, if you know of a text model that generates better responses than Gemma 2, I'd appreciate it.

I also want to clarify that I use a toaster with a 9th gen Intel Core i5 and a GTX 1660, so if you can recommend something with that in mind, and if not, then that's fine too. XD Let's try it, it's been said.

[Topper Harley (toppertrip)]

If you already know the following, excuse me: to use some models, you need to insert a key that you receive when you register on the site that provides the service.
The key is inserted in Dreamio ApiKey.

The best image generator in Dreamio in my opinion is shnell, but you get "errors" from every image generator. With errors I mean that in some cases you don't get what you want to generate.
In the image generator option you could get Together AI, FLUX (shnell) 1024 for free.

[Oleg Skutte]

Which variant of Gemma 3 did you try? Google recently released Quantization Aware Trainined versions of Gemma 3 models, which produce rezults comparable to bfloat16 precision:
https://huggingface.co/collections/google/gemma-3-qat-67ee61ccacbf2be4195c265b

I would recommend trying 12B version or above, though it may be slower than Gemma 2 9B.

[Chimal1994]

Originally posted by Oleg Skutte:

Which variant of Gemma 3 did you try? Google recently released Quantization Aware Trainined versions of Gemma 3 models, which produce rezults comparable to bfloat16 precision:
https://huggingface.co/collections/google/gemma-3-qat-67ee61ccacbf2be4195c265b

I would recommend trying 12B version or above, though it may be slower than Gemma 2 9B.

Hi Oleg, I first tried the gemma-3-27b-it version, but I discarded it because my toaster was having a hard time running it. Then I tried gemma-3-12b-it, it worked well in terms of speed, but the generated text was strange, it almost always ignored the State box and the variables that I asked it to follow, and things got worse if I activated the reasoning option.

[Oleg Skutte]

Maybe precision/quantization level was too low? Models from the collection above should work better in theory, but I haven't tested them yet.

[Chimal1994]

Originally posted by Topper Harley (toppertrip):

If you already know the following, excuse me: to use some models, you need to insert a key that you receive when you register on the site that provides the service.
The key is inserted in Dreamio ApiKey.

The best image generator in Dreamio in my opinion is shnell, but you get "errors" from every image generator. With errors I mean that in some cases you don't get what you want to generate.
In the image generator option you could get Together AI, FLUX (shnell) 1024 for free.

Hello Topper, the truth is that I haven't used anything that requires an ApiKey at the moment, I'm not a big fan of having to constantly connect to a server to generate stories XD that's why I installed the local models from day one

Maybe the APIs don't work like that and I'm missing a lot, but I understand that you constantly have to connect and receive a response

[Topper Harley (toppertrip)]

Yes, this way the servers can send you the result of something would require powerful machines. I don't even have such powerful computer to process images and text generation.
If you prefer you can register in alternative ways. I think the main one is using google one

[Chimal1994]

Update no one asked for but I'll write anyway

After downloading models (image generation) and placing them in the Checkpoints folder, causing some very serious errors in the process, I finally ended up discovering that the advanced section is not for decoration xD

I managed to install successfully, I think I'm not very sure but FLUX.1-schnell-GGUF works (Surely it's more credit to Oleg that he provided all the files, in .gguf format and that he has everything organized)

I have no idea if the .safetensors files provided by other people would work... things like clip_l.safetensors are still a very mysterious mystery to me

I liked Flux1, by the way, the generated images are very nice and they don't take as long to generate as about 4 or 5 minutes, which is not bad for the toaster I use XD

If I put flux-chatgpt-ghibli-lora.safetensors
In the folder called LoRA, will the images plagiarize the Ghibli study?

Stupidity is a privilege and I plan to use it! Jokes aside, if someone has the time and desire to explain what things you can or cannot put in DREAMIO that would be great, at huggingface.co there are a lot of models and you will feel like a child in a candy store

Sorry for the bad writing but I wrote this with passion and on the fly (maybe no one will read it except Oleg who is always attentive)

[Topper Harley (toppertrip)]

I used some models in local, but neither I have a computer with good performance, so I chose the providers.

[Chimal1994]

Originally posted by Topper Harley (toppertrip):

I used some models in local, but neither I have a computer with good performance, so I chose the providers.

You have mentioned twice that you prefer to use suppliers which can only mean one thing, you want to tell us the sad and emotional story of your potato.

Tell us, don't hold back, here we are all ears or rather all eyes

That or you are a provider of those services and you want to hook us up which would be an incredible plot twist.

[Topper Harley (toppertrip)]

you know, a potato can power a powerful AI, GLaDOS

[Chimal1994]

Originally posted by Topper Harley (toppertrip):

you know, a potato can power a powerful AI, GLaDOS

GLaDOS owes me a cake!

Note that you can perfectly ignore: I rewrite this because I don't want to fill this discussion with meaningless comments (I hope I don't send you a notification for rewriting something where I mention you and if yes, then sorry Topper)

Thanks to Portal's references I found that there is a model to make GLaDOS narrate the story XD

His voice makes your head hurt but for a Portal fan this is gold! You play DREAMIO and GLaDOS tells your adventure xD

GLaDOS TTS is called

[Topper Harley (toppertrip)]

XD I would't even touch it XD

[Oleg Skutte]

Originally posted by Chimal1994:

If I put flux-chatgpt-ghibli-lora.safetensors
In the folder called LoRA, will the images plagiarize the Ghibli study?

If that LoRA is compatible with the model you're using then yes, but you have to specify it in the LoRA field in the following format:

<lora:NAME:1>

Replace "NAME" with the file name of your LoRA without file extension.

[Oleg Skutte]

Originally posted by Chimal1994:

if someone has the time and desire to explain what things you can or cannot put in DREAMIO that would be great, at huggingface.co there are a lot of models and you will feel like a child in a candy store

For local image generation I recommend using ComfyUI because it supports a lot more models than the built-in inference engine and there is a lot more information about how to use in on the internet.

[Chimal1994]

Originally posted by Oleg Skutte:

Originally posted by Chimal1994:

if someone has the time and desire to explain what things you can or cannot put in DREAMIO that would be great, at huggingface.co there are a lot of models and you will feel like a child in a candy store

For local image generation I recommend using ComfyUI because it supports a lot more models than the built-in inference engine and there is a lot more information about how to use in on the internet.

I said that more than anything because I'm really enjoying the customization possibilities of Dreamio, I guess it has its limitations like everything in life, but it's still fantastic.

With Dreamio I feel something very similar to what I felt with Minecraft, absolute freedom

Without a doubt it became one of my favorite games.

By the way, would it be possible to expand the character limit of the status box (the one where we put which variables we want to track) in a future update?