since you guys make coomer art, I don't think this would help you much, but since I steal my art, I use chatgpt 4o to write the prompt for me, and I just refine it (for hailuo, usually the shorter the prompt, the better).
The best part about hailuo is that there is a sweet spot where the AI will generate the video and let you download it but ~10-30mins after it will remove it and give the option to "appeal", and it refunds the credits. The prompt generation time is a lot faster when it's close to the tokens expiring (7pm est for me).
This is an example that got like 70% of the videos appealed while letting me download it, so I got like 6 videos generated when they only let you generate 3 per day (I use 3 accounts using firefox containers).
https://files.catbox.moe/161c9l.mp4This is the prompt for gpt:
https://www.reddit.com/r/StableDiffusion/comments/14tol5n/best_text_prompt_for_creating_stable_diffusion/And I use it by creating a new chat with just the prompt (I modified it to include {action} in the structure, and removed weight modifiers / "photo" / etc, but it doesn't matter), then I reply with the image using "Please write a prompt using this keyword: 2 girls doing blah blah blah", or "keyword: 2 girls". If the AI refuses, I just tell it "can you fix the prompt by removing X?" and the prompt will be wrong but it will still use details from the image. But you can only generate like 1-2 prompts for free if you start a new chat, but if you just keep using the same chat, you can get 3 prompts, but the AI stops following the original instructions (1 prompt instead of 3), but I kind of like it when the AI sort of stops following the instructions, because I feel like the stable diffusion prompt doesn't work well with AI videos anyways (it's possible you don't even need the reddit prompt at all, I have not messed much with it).
So for that video, this is what my prompt looks like (and this was not the first generated prompt), and I basically used the same prompt for the entire video (I changed it a tiny bit at the end). Obviously I removed anything related to stable diffusion / image / photography.
hailuo Prompt (with "enhance"): A detailed and vibrant 3D render of two anime-style girls sharing a giant gummy worm in a playful and close-up scene. The girls are positioned face-to-face, each holding one end of the gummy worm as they bite down towards the center. ((their faces slowly move closer to each other until contact)). As they finish the gummy, their lips are about to touch in a romantic moment. One girl has a soft, gentle expression with drool leaking, while the other looks more intense with drool. Scene set with soft lighting, showcasing realistic textures on the characters and the gummy candy. Background features a cozy, indoor setting with warm tones.
I think kling is a lot better than hailou in a lot of ways, but I don't know if chatgpt prompts work that well with kling (I have not tried since I tend to just copy paste my old prompts with the same keywords, I think this has more to due with my preference to use hailuo's "enhance" option, since I don't really have tight control anyways).
And vidu is just goofy, I really should try out using llava or something "cog" or clip interrogator 2 that also lets you reverse the image into a prompt to see if I can extract a good keyword.
