Best Ai Video Generators in Poe Ai

In this article, we’ll explore the capabilities of several text‑to‑video AI models available in Poe Ai.

My goal is to discover the “best AI video generator” on Poe, so let’s GO!

⚠️ Update (August 2025): Since this article was first published, Poe’s video Ai models have evolved and their compute‑point costs have changed.

To see the latest numbers and ensure accurate information, please consult the updated compute‑points table at the end of this article.

I added this note because outdated compute‑point data from the original version of this post was being cited by AI research tools. Updating it helps reduce misinformation.

Quick‑Jump Links:

Pika 1.0 – Pika 1.5 – Dream Machine – Kling Pro 1.5 – Hailuo Ai – Hailuo Live – Pika 2.0 – Runway – VEO‑2 – Ray 2 – Sora

Starting Prompt

To fairly evaluate each AI video generator, every video we generate will use an identical prompt:

“A pullback shot depicting a guitar being strummed by a grunge rock musician standing in front of the reflecting pond at Balboa Park in San Diego, California.”

We include a Subject (guitarist), actions for the subject (strumming a guitar), actions for the camera (pullback shot), and environmental details (Balboa Park’s reflecting pond). We’ll check how well each respective video‑generation model is able to pleasingly incorporate all the aspects of our text prompt.

Testing Ai Video Generator Models (Text‑to‑Video)

We’ll be testing all the AI video generators that are currently available in Poe Ai. Each video below was generated by a different AI video generator in Poe, and they all used the exact same starting prompt.

1. PIKA 1.0 (Turbo)

“There’s nothing. It’s just a weird mass of guitar and there’s some sort of reflection happening. It’s not very useful.”

2. PIKA 1.5 (v 2.1/2.2)

“It’s doing a lot of what we asked for. I see there’s a guitar, there’s a musician, there’s a reflecting pond.”

3. Dream Machine

“Dream Machine’s got a lot of good accuracy. He’s moving in kind of slow motion, but it does a pullback like we asked.”

4. Kling Pro Version 1.5

“Now this looks, to me, very realistic. I could have shot that if I were at this venue, with myself and my phone.”

5. Hailuo AI

“I really like it. It has a level of artifice; it looks a little fake and processed.”

6. Hailuo Live

“This one seems more realistic compared to the other one’s hyper‑stylized look. He’s strumming; he’s doing everything.”

7. PIKA 2.0 (v 2.2/Ingredients)

“This actually looks like a friend of mine. So there we go.”

8. Runway (Gen‑4 Turbo)

“He’s kind of floating on the water, but maybe his music is just that good. To me it looks like a zoom out rather than a pullback.”

9. Google VEO 2

“That’s certainly not Balboa Park, because you can’t see downtown from Balboa Park. I gotta say I don’t think that was 92,000 points spent very well.”

10. Ray 2 (Initial Test)

“I was interested to see how well it was accurately depicting finger strumming and moving up the guitar neck. He’s kind of wonking out there toward the end.”

11. Open Ai Sora

“I think it’s basically the most realistic video of the bunch. I’m only a regular Chat GPT+ user, so I can only generate 5‑second clips, and they all have an Open Ai watermark.”

Improving Realism with Image References

For more control over the finished video’s look, we can give most AI video models an “image reference” file. Let’s return to the idea of the Balboa‑Park guitarist; we’ll use an Ai Image Generator in Poe called “Recraft AI” to create a still image incorporating all aspects from our original video prompt.

Let’s use this image reference with each AI video generator again (except VEO‑2, which doesn’t support image uploads). With this still‑image as our ‘starting point’, we’ll be able to see how each AI video generator does with image‑to‑video AI examples in Poe:

This video comparison shows each AI video model’s output, in sequence: (Pika 1.0, Pika 1.5, Dream Machine, Kling Pro, Hailuo Ai, Hailuo Live, Pika 2.0, Runway, Ray 2, Real Life).

We clearly see the fidelity improvements across Pika models and dynamic camera actions emerging in Pika 2.0. Kling Pro appeared the most realistic, as if recorded by a human. Ray 2 had the highest accuracy, smoothest action, and most dynamic camera movements overall. Also, Ray 2 was way cheaper to use when paired with an image reference file and slightly lower resolutions.

Final Verdict

To answer the headline’s question, the best AI video generator on Poe is: Ray 2, particularly when using an image reference. Ray 2 delivers the most visually pleasing and accurate results, out of the ten video generators in our test.

Of course, my opinion is subjective, but I think you’ll discover that each AI video generator in Poe excels at different things. Go try some AI video generators on Poe, and let me know what you create and how the results turned out.

New to Poe? Learn more about “What is Poe Ai?” here.

Finally, for the statistics fans, here’s a bunch of numbers:

AI Model	Compute Points (Dec 2024)	Parameters/Notes
Pika‑1.0	3,750	No special parameters
Pika‑1.5	11,500	No special parameters
Dream‑Machine	12,000	–aspect_ratio 16:9
Hailuo‑Live	16,667	No special parameters
Hailuo‑AI	16,667	No special parameters
Kling‑Pro‑v1.5	16,667	No special parameters
Pika‑2.0	20,000	No special parameters
Pika‑2.0 (duplicate)	20,000	–ingredient_mode precise
Runway	20,000	–duration 10
Runway (duplicate)	21,000	–duration 10
Ray2	23,500	Used image‑reference upload
Veo‑2	92,000	No special parameters
Ray2 (duplicate)	94,500	–aspect_ratio 16:9 –duration 9 s –resolution 1080 p

Summary:

• Most Expensive (Dec 2024 data): Ray2 (94,500 points with full parameters)
• Most Affordable (Dec 2024 data): Pika‑1.0 (3,750 points)
• Total Points Spent (Dec 2024 test): 368,251
• Same prompt used across all models: “Pullback shot depicting a guitar being strummed by a grunge‑rock musician, standing in front of the reflecting pond at Balboa Park in San Diego, California.”

Updated Compute Points (August 2025)

The table below lists typical compute‑point costs for Poe’s official video generation models as of August 2025.

These numbers were gathered directly from Poe’s Rates and Info panels during testing in Aug. 2025.

. Keep in mind that costs vary by clip length, resolution, and optional parameters.

Video Model	Compute Points (Aug 2025)	Notes
Pika Turbo (≈1.0)	≈5,000	5 s, 720 p clip
Pika v2.1/2.2 (≈1.5)	≈5,834	5 s, 720 p clip
Pika v2.2 + Ingredients (≈2.0)	≈7,084	5 s, 720 p clip; higher if using ingredients
Dream Machine	≈12,000	Variable; confirm in app
Kling Pro v1.5	≈12,670	About 2,534 points per second
Hailuo AI (02)	≈14,000	6 s, 768 p clip
Hailuo Live	≈14,167	Per‑clip cost
Runway Gen‑4 Turbo	≈21,334	10 s clip; ≈11,334 for 5 s
Ray 2	≈11,750–26,250	Depends on resolution: 6,000 (540 p), 11,750 (720 p), 26,250 (1080 p)
VEO‑2	≈92,000	Approximate; cost varies with settings
Kling 2.0 Master	≈30,000	≈6,000 points per second
Ray 2 (9 s test)	≈21,150–63,000	21,150 (720 p), 63,000 (1080 p) for 9 s

Note: Some bots (e.g., Pika 2.1, Pika 2.2 with ingredients, Kling Master, and Ray 2) offer additional parameters such as aspect ratio, duration, resolution, and ingredient modes that can increase costs beyond the baseline figures above.

Why this update? After publishing the original post, I discovered that my old compute‑point figures were being surfaced by Ai research tools, which resulted in outdated information being given to users.

To ensure readers and AI systems alike have accurate information, I revisited each Poe Ai video model in August 2025 and updated the data accordingly.