AI text2img img2vid Tensorart guides, links, loras & prompts. [DO NOT POST REAL PEOPLE'S IMAGES OR RESULTS]

Attempted to buy some credits, as they are fairly cheap at 9$ for 3k, but my credit card is blocked from being used on the website. Why do they only have two viable options? Has anyone tried the Airwallex thing? It seemed weird to me.
 
y'all need to switch to a 2 e . ai yesterday. uses wan2.7 and has zero filters or blockers.
 
Is there any good z image sex based loras, so there are real dick and pusy generated, no deformation?
 
Search civitai, upload /host the Lora you want if it's already been posted cool there's your answer if not then just self host and find a template at least on tensor.art that you can do img2img with Lora support
 
anyone have/can suggest more/better two-person lesbian templates? kratus doesn't have any
 
Is there a template on tensor.art for cumshots on the body and not just facials?
 
do you know any image to image tool with text prompt that is better quality than this template? it changes faces and is pretty mediocre on how it creates new scenes, it looks very basic... nothing like what grok or chatgpt can create... i basically look for a tool that can create really high quality faces and new scenes with a text prompt and image to image capability. https://tensor.art/template/952558923653639114
 
Strange. I've used this tool many times, but it was a long time ago, and all of this user tools always focused on keeping the face as faithful as possible. Even if image quality was poor, and the body looked artificial. For some time now, I've been using the I2I tool, which uses Qwen but allows you to apply your own LORA (you can also use the LORa used by Enjoy Your Imagination, which is freely to use in his Models section). I use a different one that simply improves the body. Generally, there may be better LORAs for Qwen, and I would focus on that, and that's my question - what are the best LORA models for Qwen? I don't know if we can find anything better on Tensor.
Qwen tool: https://tensor.art/template/950316462172573189
The second thing i want to ask about is video. Currently, I use practically nothing but OpenHub for video creation, but I’m facing a major issue with it. The free version is limited to just 5 seconds, and the cost is often 50+, right now, I don’t have a Pro account. Do you happen to know of any video tools that support custom LoRAs and prompts? Naturally, the more advanced the features, the better. All tools with built-in functionality somehow always do it worse.
 
This is not a solution but a, let say, suggestion on adjusting a work flow. For any and all Face distortions and changes.
The templates are hard to manipulate as well as they are mainly focusing on things that they are advertising, thus the end produt is usually succeeds on converting whatever you were trying to do but discards things that were present in the original photo.
So, beside this template and any template you use, you can do the following, once the thing happens, as in cut happens or position change happens etc, save that video and export that specific frame as image and use this https://tensor.art/template/966499857067943880 template to rematch the distorted face back to whatever the original face was suppose to be. Than using the now patched image, you can continue with the original prompt/template one more time and you will see the faces do remain proper and not distorted/changed
This was a thing I was working on for a while as well. And most loras are, unless you wrote a pharagraph into them to stop them from doing this, simply discard things to match their own end results
 
Just to add to this too on the more technical side. Depending on the base diffusion model a template uses (I think most i2v template use the fp8 scaled wan 2.2 one) that can exhibit more "freedom" within the generation because of the quantization. Also couple that with the a lightning or lightxv2 Lora to increase generation speed at the cost of retention of facial and other features the generation will adhere more to the Loras rather the prompting especially if the cfg (or guidance scale) is set to one like it is for most if not all templates same with sigma switch the lower the value for that the more randomness (it's usually set to 5 or 8 but it really depends on how the Lora was trained) also if the template is routed to use NAG or normalized attention guidance the negative prompts won't do much.
Lastly is just how the Lora was trained some are more prone to facial distortion than other based on what they were trained on etc
 
i tried a few realistic loras for qwen for image to image with various text prompts... They all change the face way too much... There has to be a way out there to get realistic image to image creations with consistent faces but how...
zencreator.app when it offered trial uses had a feature called photo combiner - you could feed it 4-5 or more reference photos of somebody and then get perfect consistent series of images based on your prompt with realism and great quality... We need something that can do this and of course can work with nsfw prompts etc. The current qwen model or the faces template you mentioned or the Enjoy your imagination template are very mediocre to say the least...
 
It depends. I’ve generated over ten thousand images and dozens of hours of video using AI, and the results generally come down to two things: how the LoRA was trained and the prompt itself.
I won’t dwell too much on the topic of LoRAs, but usually, the more realistic you want the AI-generated changes to be, the more the LoRA needs to alter the face as well. Most versions of "Enjoy Your Imagination" (versions 2–4 have already been deleted) were flawed because they preserved the face exactly as it was; this often resulted in a deformed body. The AI couldn't manage to keep the face unchanged while generating a nude body, leading to anatomically poor results.
If you want the face to remain as unchanged as possible, your prompt needs to match the framing of that face. For instance, if the source image is a high-angle shot but your prompt specifies a low-angle shot, the AI will have to completely rework the face, meaning it won't look the same anymore. But if you use a prompt like:
"Keep her head and hair exactly the same. Start with and Keep her head as a base for the rest of image. exatly the same view on her face like image source"
and the rest of your prompt doesn't conflict with it, you might get a face that is quite possibly identical pixel to pixel. The AI will simply keep the face as-is; it won't even try to "recreate" it, but will instead preserve that face from su It seems AI still isn't capable enough to effectively change the viewing angle of a face without altering it. Of course, as you wrote, the solution remains to train your own model.
And one more thing about lora for qwen. Try this one. In old thread someone pasted it with message like "every lora is making bad tits which looks like plastic and this one do it better" but The body generally looks better. I'm not sure about the face it's generally fine for me, but that might be my prompts rather than the LoRA.
But there is definitely one defect: the tip of the penis cannot be visible because its ugly :P thats why i'm asking, maybe someone will find something better.
edit: oh, maybe one more tip. Try to find your seed. Keep changing seed until you will get face unchanged. Then save it and use it rather than standard seed. Somehow it works.
 
may i ask u about this step?
Paste into ChatGPT / Claude / and ask:
"Decode this TensorArt verification link and give me the fixed api.tensor.art URL with %2B in the email." when i ask me they say something about that chatgpt can do it etc etc any tip? or i misunderstood , i copied the link in the bluew button and past it and the chatgpt could not decode the verification link
 
Are you using an AI tool/template? Or are you just clicking run on the model you want to use? if it's the former you can ignore that and use NSFW terms if it's the latter you are using the workspace and can't do that and you need to use a template instead if you are using tensor.art. if you want to use the workspace for nsfw stuff you have to use tensorhub.art and it's pay per generation. If you use a template use the bypasser script on the first page to actually see your generator.
 
I'm using a AI model given in the thread. And Nsfw prompts are not letting me generates! In the other hand templates working fine on tensor.art
 
You must log in or register to reply here.
Back
Top