Hi

Let’s make a place where people can share tips and advice, or ask for help with creating AI art. I’ll start with non-local generators.

Advice

“I love this image! How do I generate something like this?”

Use AI to analyze images you like and figure out the style, lighting, etc. you want.

But please don’t upload anything you don’t own the rights to onto platforms that may use uploaded images for model training or other purposes. Respect the artists you like.

Free generators can surprise you

Prompt is most important, not the model.

Even sinple, outdated but free tools like Perchance can generate decent characters. Don’t give up too quickly. Sometimes it takes many generations to get a good result.

Grok Imagine on grok.com currently offers 1–2 image generations or edits per day, plus one 480p, 6-second video. The video quality is surprisingly good if you use a vertical aspect ratio. 16:9 gets badly affected by the low resolution and compression.

Grok also has a much-needed toggle that lets you turn off training on your uploads.

Grok has recently become very good at image editing, too. It’s useful for fixing AI glitches in an otherwise good image.

If you can’t afford paid services, look for websites that give you a few free credits when you start. Some also let you earn more through daily logins. That should be enough to test a better image model or sometimes a decent video model, although usually with very low resolution and quality and a maximum of 4–5 seconds.

VIDEO

First of all, if you want decent quality from advanced models, prepare yourself for the prices.

A 10–15 second 720p video can cost $2–3 per generation. Usable settings usually start below $1 and go up to around $3 (for like 720p 10-15s). If you want 30 seconds from the newest models, it can be as much as $25-40 for a single video.

My experience with video models

GROK – Probably useless for most of the content you see here. It’s strictly moderated, and you pay for an attempt even if you receive nothing. The recently introduced weekly quotas also allow only a few videos per week. Let’s say 3–10 if you want 16:9 with quality above YouTube’s 240p level. That’s on the $30/month tier.

Test the free version first to see whether it allows you to generate what you want. It’s a very good option for SFW content and probably horrible for NSFW. If you see something made with Grok Imagine that is even slightly NSFW, you’re most likely looking at something generated a month ago. Voice is rather bad in general. Decent for dancing.

Kling 2,x and 3.0 - Can be very good or very bad, depending on what you dislike. For me, it was mostly bad, so I don’t have much experience with it. Only the 1.6 model renders the style I like, but everything besides the image style is basically broken. Newer models heavily favor realism, which is exactly what I hate, so I unsubscribed and sadly can’t share much about this model.
Voice is much better than Grok, with lots of "acting" and correctly estimated emotion.
You can hear the good voice acting on my video here: https://e6ai.net/posts/178827
Notice how the character adds emotion to the words and adjusts the tempo and tone.
Believe it or not, my direction only covered about half of what you can hear here. Kling did the other half on its own, which is very impressive. Kling isn’t that expensive either if you use their own website directly. AI aggregators can add markups of 300% or more.

flaws?
May be bad at dancing, but I probably just haven’t tried enough to figure out how to prompt it properly.

Seedance 2.0 Fast - The cheapest of the good models. Around $0.70- $1.30 per video that doesn’t look like crap. If you choose the provider wisely, it’s fairly permissive, although I’ve never tried generating porn or anything really explicit.

As with the Seedance 2.5 models, unfortunately it skips frames, so there’s a lot of judder.

GENERAL ADVICE

Don’t know? Ask a chatbot. They can be surprisingly helpful with anthro-related questions. Usually, though, what looks like the perfect solution simply doesn’t work. But it might work with another AI model, or with the same model in the future, since models keep changing.

Have a very good memory or MAKE NOTES. Create a file and save the things you found that work. Generating with paid services is expensive. Don’t waste money because you forgot to mention something you solved month ago. Use low-quality settings for previews first, and generate the real thing once you’ve got it right.

What I’d like to find answers for


How to achieve an unrealistic fur style in 3D semi-realistic images and
How to avoid bias towards realism.

I want to generate anthro characters from another world, with perfect, clean, fluffy fur - not something that looks like real fur, and definitely not something that looks like a real animal. Because of this, 99% of the Kling videos I got are not what I want, and it’s a struggle with Seedance and Grok too.

How to prompt for delicate, bushy but smooth, combed, “perfect-looking” fur?
And another problem - Grok tends to produce a pine-cone look, with strands that look like a pine cone. I’ve tried everything I could think of, and it’s a miracle to get anything that isn’t this, or something else like extremely long strands, especially strands that look tangled or not perfectly clean. I’m aiming for a vibrant, almost glowy look, and the fur should look very delicate.

How to keep anthro as anthro

Is there any way to stop a specific AI generator from changing my anthro character into a four-legged feral form? I’ve lost more money on this problem than I’d like to admit.

Tail motion.
I’ve gotten a few good results, but the number of ideas I’ve already tried to get the tail motion I want is easily above 100, and I still don’t have a solution. What worked on Grok half a year ago no longer works. Other models have different prompt interpretations too.

I’m not a native English speaker, so I’d be very grateful if someone could tell me a working way to get a flattened-U-shaped motion path correct every time.

Stutter in Seedance videos

How to fix the stutter in already-generated Seedance 2 videos? I generated one prompt 20 times and only got two good videos. One is 10/10, and I’d really like to fix the stutter on that one. Even Grok Imagine often drops frames when the image is complicated.

"perfect fur style"

My preferred style is “unreal, perfectly beautiful.” It can be glowing. If anyone has a working idea for making all the fur strands glow without realistically lighting the surrounding areas, I’d be more than grateful. That unfortunately ruins the contrast between the glowing white and dark black fur patches.

Usually, the problem is that the glow doesn’t cover the entire character or gets lost when the character changes pose. Some models have improved this recently, and I do include text about it in the video-generation prompt, but there’s no silver-bullet solution to this - at least none I’ve found so far.

Grok voice

How do I get Grok to use a female voice that doesn’t sound robotic, and how do I make it stop wasting money by suddenly giving a female character a male voice? I’ve only gotten good results by accident, and only a few times.
I tried learning how the various female voices are called in English and Grok simply ignored everything in the prompt.

Dance matching music

Any ideas on how to keep a character dancing while staying on rhythm, so that if I edit in music with the correct tempo, the movements match the beat? Or how to make video editing easier for this task? Preferably with free or cheaper video-editing software.
I can’t afford the best apps at ~$300/year plus $300/year plugins, and I don’t have a decent PC.

Keeping the same voice for the character without using an audio reference.
If you've found a good method for a particular AI, please share it!