AI SEO Weekly

[{"selector":"#anim-6c13e477-16cf-4fb7-b716-e40d29bbb3f6","keyframes":{"opacity":[0,1]},"delay":2000,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] By @ValeIzzo86 & @CyberAndy

2015

[{"selector":"#anim-ddf14b7e-6cb4-47ce-9f95-a378437c7ca2","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] [{"selector":"#anim-33e6fba8-9a0d-4d3a-be7c-16df664a66a1","keyframes":{"opacity":[1,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-42841f82-939a-4cb3-b8d9-4658b2c907bf","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] Watch the video

IMAGE

[{"selector":"#anim-fdae4a2b-e85e-404d-a194-b3d4635711bd","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] Curved Arrow

What if you flipped that process around?

[{"selector":"#anim-c43ad32f-c3ce-46f3-8840-78b5a739f180","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}]

PROMPT

[{"selector":"#anim-9913c489-ae19-436b-a1e5-80b5e0be51f5","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] [{"selector":"#anim-6e1aa203-fd60-4870-9055-42b61eeea9af","keyframes":[{"transform":"scale(1)","offset":0},{"transform":"scale(1.5)","offset":0.33},{"transform":"scale(0.95)","offset":0.66},{"transform":"scale(1)","offset":1}],"delay":0,"duration":1450,"easing":"ease-in-out","fill":"both","iterations":1}] Curved Arrow

WHAT IS A DIFFUSION MODEL?

[{"selector":"#anim-97b497aa-2608-431e-9334-097b61990dba","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] [{"selector":"#anim-3ef3f2b6-59be-4bf3-9a20-169473c55d49","keyframes":{"opacity":[1,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-3f208591-83b2-44b6-aaac-da52bf4b3e42","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Diffusion models are like autoencoders, which transform input data into an embedding representation and then reproduce the original data from the embedding information.

[{"selector":"#anim-3f3c0558-1fee-4b3c-92da-c553c53744e1","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}]

ORIGINAL GIOTTO

[{"selector":"#anim-422da59f-fa89-4302-9d9f-5289889a66b8","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] Curved Arrow

A new AI System - Converts text prompts into images. - Maintain semantic consistency in the images it creates. - Can generate multiple variations of the same image. - Can edit an existing image.

[{"selector":"#anim-9ef3dd84-05bc-46f1-8d61-6d42e0660c2b","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}]

WHAT IS DALL·E 2 ?

[{"selector":"#anim-126f90ad-d324-44e5-91a4-055c238e6697","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] [{"selector":"#anim-10b5e946-ff88-4705-85c1-7ee796b7f267","keyframes":{"opacity":[1,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-972b0851-3459-4a39-b8a6-75bf5eb6044c","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] Learn More

WHAT DOES IT MEAN FOR SEO ?

[{"selector":"#anim-991953f6-c117-44cd-851d-2afe3e525fc2","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] [{"selector":"#anim-85e479c1-4474-4bc7-b9a1-14ecb36ceea3","keyframes":{"opacity":[1,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-b02bad33-1787-4b26-be81-83ba82b5f0b1","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-f6ce9483-c226-443e-b6e3-06b1e192f4c7","keyframes":{"transform":["translate3d(-114.75903%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .8, 1)","fill":"both"}] [{"selector":"#anim-108e1e4c-9d80-414d-86af-fa4b5b990372","keyframes":{"transform":["rotateZ(-180deg)","rotateZ(0deg)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .5, 1)","fill":"forwards"}] [{"selector":"#anim-b14cece0-941b-437f-a774-a7678a0fb3f1","keyframes":{"transform":["translate3d(-115.55555%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .8, 1)","fill":"both"}] [{"selector":"#anim-a28e2d8e-fb0b-4784-8303-90913227eb33","keyframes":{"transform":["rotateZ(-180deg)","rotateZ(0deg)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .5, 1)","fill":"forwards"}]

If we have data (let’s say in a Knowledge Graph), we can convert structured data into natural language.

[{"selector":"#anim-915021a8-1549-493a-84e5-62c3b8cc2b73","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] Read More

You can train your model to write, for example, the recipe for Amatriciana, using triples in your knowledge graph.

[{"selector":"#anim-02337f76-4ba4-47fa-8543-93357ae1121a","keyframes":[{"transform":"scale(1)","offset":0},{"transform":"scale(1.5)","offset":0.33},{"transform":"scale(0.95)","offset":0.66},{"transform":"scale(1)","offset":1}],"delay":0,"duration":1450,"easing":"ease-in-out","fill":"both","iterations":1}] Read More

More importantly you can create product descriptions where they are missing.

[{"selector":"#anim-df41e703-b2e0-479e-a4e5-69372c01132c","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] Read More

Can we make our prompt even better by combining image with text generation?

[{"selector":"#anim-cbb3649f-81f2-4968-a4de-a7623af25b7b","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] [{"selector":"#anim-66877391-fe5f-4265-9749-8583e6d50214","keyframes":{"opacity":[1,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-30eab0fb-d078-4a06-a557-4a47ede2279f","keyframes":{"transform":["scale(3)","scale(1)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-b245738f-9505-4cef-a01f-07630199434c","keyframes":[{"transform":"scale(1)","offset":0},{"transform":"scale(1.5)","offset":0.33},{"transform":"scale(0.95)","offset":0.66},{"transform":"scale(1)","offset":1}],"delay":0,"duration":1450,"easing":"ease-in-out","fill":"both","iterations":1}]

We basically evaluate the quality of the completion (the AI generated product description) by generating an image with DALL·E 2. Then we compare the visual similarity between the generated image and the real photo of that product to see what completion works best.

[{"selector":"#anim-a0437d4b-7662-4e5d-b819-d033e4688859","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}]

ORIGINAL IMAGE

[{"selector":"#anim-4d0095e8-2d64-4aac-837e-d4f3dea1efae","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}]

We compare synthetic images with the real product image to evaluate the best prompt.

[{"selector":"#anim-5a32155a-a9e1-4e61-b6ea-6240816cb53c","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}]

[{"selector":"#anim-805b7c16-3c09-40ad-a67e-80ef382d3a5a","keyframes":{"transform":["scale(0)","scale(1)"]},"delay":500,"direction":"normal","duration":1401,"easing":"cubic-bezier(0.77, 0, 0.175, 1)","fill":"both"}] Contact us

AI SEO Weekly

Prompt engineering in SEO

2015

2015

Automated images captioning

Automated images captioning

Machine learning algorithms were already able to label objects in images, and now they were learning to convert those labels into natural language descriptions.

IMAGE

IMAGE

CAPTION

CAPTION

What if you flipped that process around?

PROMPT

PROMPT

IMAGE

IMAGE

WHAT IS A DIFFUSION MODEL?

WHAT IS A DIFFUSION MODEL?

DIFFUSION MODELS are generative models that learn to create images by gradually noising and denoising their training samples.

Diffusion models are like autoencoders, which transform input data into an embedding representation and then reproduce the original data from the embedding information.

ORIGINAL GIOTTO

ORIGINAL GIOTTO

FAKE GIOTTO

FAKE GIOTTO

A new AI System - Converts text prompts into images. - Maintain semantic consistency in the images it creates. - Can generate multiple variations of the same image. - Can edit an existing image.

WHAT IS DALL·E 2 ?

WHAT IS DALL·E 2 ?

DALL-E 2 is a new AI system that can create realistic images and art from a description in natural language.

WHAT DOES IT MEAN FOR SEO ?

WHAT DOES IT MEAN FOR SEO ?

👉 It helps us understand how Google Lens works

👉 It helps us understand how Google Lens works

👉 It is precious to generate content both textual and visual

👉 It is precious to generate content both textual and visual

If we have data (let’s say in a Knowledge Graph), we can convert structured data into natural language.

You can train your model to write, for example, the recipe for Amatriciana, using triples in your knowledge graph.

data-to-text

data-to-text

More importantly you can create product descriptions where they are missing.

Can we make our prompt even better by combining image with text generation?

Can we make our prompt even better by combining image with text generation?

Yes we can!

Yes we can!

Here are a few experiments that we’re running to improve the quality of our prompts.

We basically evaluate the quality of the completion (the AI generated product description) by generating an image with DALL·E 2. Then we compare the visual similarity between the generated image and the real photo of that product to see what completion works best.

ORIGINAL IMAGE

ORIGINAL IMAGE

Synthetic Images (generated using DALL·E 2)

Synthetic Images (generated using DALL·E 2)

We compare synthetic images with the real product image to evaluate the best prompt.

MORE SEO STORIES

SEO IMAGE OPTIMIZATION

GPT-3 FOR PRODUCT DESCRIPTION