I Tested the Best Artlist AI Video Models - Part 2

❤️ Before we get started I'd like to thank you for using my affiliate links to sign up to free trials, LLMs are constantly stealing my content and you help me stay afloat and create more of this content to AI enthusiasts and small business owners. ❤️

In part 1 I put six Artlist AI video models through two tough prompts: a complex multishot scene with text rendering, and a breakdancing nun. Sora 2 won the nun, Veo 3.1 won the robot handshake. If you missed it, catch part 1 here.

In this post I'm going deeper into what the Artlist AI video models can actually do when the prompts get more creative. Prompt 3 is about cinematic aesthetic and scene-building — can a model create something that's genuinely beautiful to look at? Prompt 4 tests start and end frame consistency, which means I give the model a first and last frame and it has to build a coherent transformation between them. That second one is brutal and the results show it.

Same rules as before: first generation only, no prompt tweaking.

Prompt 3: The Origami Birds

The prompt: Animate a serene scene where a woman in a flowing white lacy dress with intricate detailing stands confidently in front of a floor-to-ceiling glass window that reflects the peaceful garden behind her. Vibrant origami birds cascade gently out of her head. The origami swallows swirl around her in slow, graceful movements, like a school of birds would do in nature in beautiful patterns together. They fly into a spiral around her and then away all at once. Add soft natural light that highlights her features.

What I was testing: aesthetic appeal and cinematic motion.

This prompt is less about technical accuracy and more about feel. Can the Artlist AI video models create something genuinely serene? Can they handle coordinated flock movement that looks natural rather than glitchy? And for the models with audio, can they improvise something that fits the mood?

🥇 Sora 2

Sora 2 is my favourite for this scene. The garden reflecting through the glass window is genuinely serene, the lighting is soft and flattering, and the ambient audio complements the mood without feeling forced. The birds don't follow the flock choreography perfectly — they dissolve slightly and fly a bit too close to the subject — but the overall composition is beautiful enough that it doesn't matter much. This is the kind of output you'd actually use.

🥈 Runway 4.5

Runway is a real surprise here and deserves the second spot. The aesthetic is completely different from every other model in this test — it has its own visual language, almost like it's operating from a different creative reference point entirely. The scenery and overall composition are strong. The one thing missing is audio, which would have been genuinely interesting to hear on this particular prompt given how well the visuals landed.

Kling 3.0

Kling does an acceptable job but doesn't fully deliver on the brief. The bird movements don't match the flock choreography described in the prompt, and the background reads as a plain window rather than a glass surface reflecting a garden. The scene works well enough but lacks the aesthetic depth the prompt was going for.

Veo 3.1

The background garden is well executed, but the birds land on the subject rather than swirling around her, which breaks the serene tone immediately. The overall scene doesn't have the delicate, ethereal quality the prompt calls for. Technically competent in parts, but the mood doesn't land.

Wan 2.7

Wan defaulted to an animated style again without being prompted to do so. The bird movement and background are actually well handled, which makes the animation choice more frustrating — the core motion is there, but the aesthetic kills the serenity the prompt is built around.

LTX 2 Pro

The birds look unfinished and the overall scene doesn't read as serene. The SFX feel out of place rather than complementary. LTX struggles here both on motion quality and on atmosphere.

Prompt 4: The Space to Hawaii Transformation

The prompt: The woman stands facing the camera in a space suit. The background is pulled away and is replaced by a sunny Hawaiian background. She spins around once and when she faces the camera again she is wearing the Hawaiian dress. The hibiscus flower and flower garland floats through the air and lands on her.

What I was testing: start and end frame consistency.

This is the prompt that separates the Artlist AI video models most clearly. Worth noting upfront: not all of them support end frames. Sora 2, LTX 2 Pro and Runway 4.5 don't have that feature, so those three were working from the start frame only. That's a significant disadvantage for a prompt that's specifically built around transformation.

🥇 Kling 3.0

Kling wins this one, and it's not just because the others struggled. The background transition is handled creatively — rather than pulling away as prompted, the new Hawaiian scene grows into the frame organically, which actually works better visually. The spacesuit transforms into the dress smoothly and naturally. The flowers grow and move with a subtle elegance. The spaceship in the background was completely unprompted and its SFX appeared on their own, which is a nice bonus. The garland and flower appear during the spin rather than floating in afterwards, but honestly no other model gets close enough for that to matter.

Veo 3.1

Veo slices the frame apart to reveal the Hawaiian scene underneath rather than pulling the background away, which is a different interpretation but not an unworkable one. The spin happens but the garland ends up being thrown rather than floating gently. The flower appears next to the subject in a slightly odd way before settling. Audio is missing again, which is becoming a pattern with Veo on these prompts. Not impressive for start to end frame work, but not a complete miss either.

Wan 2.7

Wan's output has its usual slight animation lean. The background rolls out rather than pulling away, and the subject has to spin twice before the transformation completes. The hibiscus flower appears oversized and the subject briefly disappears inside it before it fades out. Not quite there, but the transformation logic is at least present.

Sora 2

Without end frame support, Sora improvises and the result is weak compared to its other prompts. The limbs look awkwardly cropped, the dress is inconsistent across frames, and the background transition feels unresolved. The flower waits in the corner of the frame before drifting across, which looks more like a placeholder than a design choice. This is clearly not Sora's strength.

LTX 2 Pro

LTX also has no end frame support and it shows. The post-transformation subject doesn't maintain visual consistency with the start frame. Beyond that, the Hawaiian background doesn't appear, the garland is loosely interpreted at best, and the model misses several of the basic requirements of the prompt, not just the details.

Runway 4.5

Runway gets a lot wrong here. After the spin the subject's appearance changes significantly, which breaks character consistency entirely. The background slides sideways rather than pulling away, and the Hawaiian volcanic landscape never appears. The dress has printed flowers rather than the garland, which does eventually make it into the frame. Not usable for this type of prompt.

Which Artlist AI Video Models Are Worth Using?

After four prompts across six models, here's where each one actually stands.
Sora 2 is the most well-rounded model for complex scenes with crowds, atmosphere and audio. It handles human motion better than anything else here and its first generations are consistently the most usable. If you're creating content where mood and realism matter, start here.

Veo 3.1 is the strongest for technical accuracy and text rendering. It came closest to hitting every requirement in prompt 1 and would have been near-perfect if the audio had worked. The audio issue is a real frustration given how consistent the visual output is.

Kling 3.0 is the model to reach for when start and end frame control matters. It handles transformation prompts better than the rest and has a cinematic quality when given the right brief. Less reliable on complex motion, but its strength in structured transitions is no surprise.

Runway 4.5 has a genuinely distinctive visual style that stands out from the rest of the field. It surprised me on prompt 3 with its aesthetic, and it's worth using when you want something that looks different. Audio and multishot are its main gaps right now.

Wan 2.7 has real strengths in text rendering and background detail, and it added multishot spontaneously on prompt 2 which none of the others did. But the automatic animation style is a persistent problem that limits how useful it is for realistic content.

LTX 2 Pro struggled across most prompts. It has occasional flashes of interesting output, the black and white on prompt 2 was a good accident, but it's not consistent enough to rely on for first-generation results.

When testing the Artlist AI video models across all four prompts, I honestly didn't expect Sora 2 to hold up this well against newer models. It won two out of four prompts and was competitive across the board. That would likely look different if I could have included Seedance 2.0, but the human content restriction made that impossible for this test.